What this is

Answers Jessica’s Point 5 — “comparison of direct mechanism of action between plixorafenib and other competitor BRAFi or RAFi, biochemical structure, resistance mechanisms” — against the seven molecules she named and the axis she specified: binding pockets, monomer versus dimer, paradoxical activation. It audits every paradox-activation claim in the set: who makes it, in which venue, and what the primary evidence actually supports. The point is where those two come apart.

Feeds FORTE Q7 (GO/NO-GO) — is monotherapy positioning defensible against competitor BRAF/MEK combinations? The paradox argument is the axis that question turns on.

Why it matters

Plixorafenib’s differentiation rests on paradox breaking. That argument is only as good as what the competitors can and cannot evidence — and, uncomfortably, as good as whether binding class predicts paradox behaviour at all.

How claims are graded

label and peer are grounding. conference is provisional. company and review are awareness only and can never establish that a molecule does or does not cause paradoxical activation.

How it was built

Seven agents, one per molecule, each then adversarially re-verified by a second agent prompted to refute rather than agree — attacking venue integrity first. 133 corrections were applied before synthesis.

Scope

Public domain only. Plixorafenib is excluded by rule — not researched, not inferred. The matrix leaves the column for Fore to complete from cleared internal data.

Not SME-reviewed. Every [FLAG:] item needs human confirmation before external use.

Molecules

early-gen type I, paradox breakers, pan-RAF

Key findings

claim vs evidence

Corrections

133

applied in the refutation pass

Unknowns

recorded, not smoothed over

Open flags

human verification required

01Key findings

NO MOLECULE IN THIS SET CAN CREDIBLY CLAIM PARADOX BREAKING IN ABSOLUTE TERMS ON THE PUBLIC RECORD. Two claimants (claturafenib, NXP200) use explicitly RELATIVE language in their own peer- and conference-facing venues ('less paradoxical activation than encorafenib'; 'significantly less paradoxical activation… spares wt-BRAF-containing RAF dimers') and thereby AFFIRM residual paradoxical activation. One (mosperafenib) is contradicted by its own sponsor's readable supplementary data. One (tovorafenib) has an approved-label NF1 warning describing MAPK activation. The absolute formulations exist only in company-tier and review-tier material.

THE STRONGEST CLAIM ANY COMPETITOR CAN MAKE IS 'ATTENUATED, IN A DEFINED CONTEXT, AT A DEFINED TIER' — and that formulation is available to only two molecules. Tovorafenib: true as scoped to BRAF fusions (label + peer + regulatory-assessment), false as generalised to RAS-pathway-activated BRAF wild-type settings. Claturafenib: relative versus encorafenib in a 19-line panel, one sponsor paper, one hour, no RAS-mutant model, magnitude never reported.

MOSPERAFENIB'S OWN SUPPLEMENTARY DATA REFUTES ITS OWN ABSOLUTE CLAIMS. Suppl Fig S2 of Mol Cancer Ther 2026 shows pERK rising to ~3.65x DMSO in a BRAF wild-type / KRAS-mutant line at concentrations inside the compound's own therapeutic window, with the bell-shaped signature of paradoxical activation — against congress language of 'not triggering MAPK over-activation in BRAF WT contexts AT ALL LEVELS'. It is also not uniformly below encorafenib (SW480 at 1000 nM, both blocks). And the asset was REMOVED FROM ROCHE'S PHASE I PIPELINE as of 23 July 2026, reason undisclosed.

THE ENTIRE NXP200 PARADOX CASE IS ONE HCT116 EXPERIMENT WITH THE ASSAY NOT NAMED. No dose-response, no quantification, no second RAS-mutant line, no keratinocyte or skin model, no in vivo paradox readout, one comparator. There is NO label-tier and NO peer-tier evidence for this molecule anywhere, both verified by retrieval. The chemical structure is not even public. The verb escalates from 'did not induce dimers' (conference) to 'physically disrupting the formation of the BRAF dimer' (company) with no experiment closing the gap.

THE CLAIM HARDENS AS IT TRAVELS — THIS IS THE MOST RELIABLE TELL IN THE DATASET, AND IT IS THE STRONGEST SINGLE LINE OF ATTACK. Holderfield 2014's 'It has been speculated that…' becomes a flat assertion downstream. Sun 2017's explicit subtherapeutic-dose qualifier vanishes in the review literature. Roche's patent language ('reduced by more than 25%' vs dabrafenib — i.e. every compound still induces paradox) becomes 'at all levels'. Claturafenib's 'less paradoxical activation' becomes a blog's 'significantly sparing normal cells'. In every case the absolute version is the one that reaches a competitive deck.

ENCORAFENIB'S OWN LABEL, IN BOTH JURISDICTIONS, DESTROYS ITS 'HIGHLY SELECTIVE' FRAMING: BRAF V600E 0.35 nM, WILD-TYPE BRAF 0.47 nM, CRAF 0.30 nM. CRAF is inhibited MORE potently than the mutant target. No structure, no docking, no inference required. Vemurafenib's USPI likewise says it inhibits CRAF, ARAF and wild-type BRAF 'at similar concentrations'; its SmPC gives V600E 10 nM vs WT BRAF 39 nM. The biochemical substrate for paradox is disclosed by the regulators themselves for all three early-generation agents.

BINDING CLASS PREDICTS NOTHING AND THE STRUCTURAL LINE OF ARGUMENT SHOULD BE RETIRED, NOT CAVEATED. Tkacik 2026 (eLife RP110344) reports that type I AND type II inhibitors induce paradoxical activation of all three RAF isoforms, with the type II inhibition phase suggestive of POSITIVE cooperativity — in conflict with the prevailing negative-allostery model. Any 'molecule X is class Y, therefore paradox / no paradox' argument is refuted at the class level.

A LARGE SHARE OF THE FOUNDATIONAL 'PARADOX' LITERATURE IS TOOL-COMPOUND LITERATURE, AND THE SUBSTITUTION IS ELIDED DOWNSTREAM. Hatzivassiliou 2010, Heidorn 2010, Sievert 2013 and Poulikakos 2010's definitive chemical-genetic panels all used PLX4720 or other tools, not the marketed drug. Vemurafenib was NEVER TESTED in Sievert 2013 — which nonetheless treats its paradox as 'the well-described paradoxical activation', an unexamined cited premise. Dabrafenib has NEVER been tested in any BRAF-fusion model.

BOTH LABEL-TIER PARADOX WARNINGS (vemurafenib 5.2, dabrafenib 5.2, encorafenib 5.2) ARE CLASS STATEMENTS ABOUT 'BRAF INHIBITORS', NOT MOLECULE-SPECIFIC FINDINGS, AND NONE CITES A STUDY. A class statement in a label is a supported LABEL claim, not a molecule-level experiment. Conversely, the labels DO operationalise the paradox into hard prescribing rules — wild-type-BRAF non-indications cross-referenced to the warning, and permanent-discontinuation rules for RAS-mutant non-cutaneous malignancies.

VENUE MUST BE SPECIFIED DOWN TO THE JURISDICTION. Encorafenib's EU SmPC contains NO paradox statement at all, while its FDA label carries the full 5.2 warning — so citing 'the label' without saying which one yields a false negative. Vemurafenib: 'paradox' appears ZERO times in the entire EMA SmPC and 'dimer' nowhere in the USPI. Tovorafenib: neither approved label uses the word 'paradoxical', yet both describe the phenomenon in the NF1 sections, and the explicit regulator statement that 'paradoxical activation could be an issue' sits in the CHMP assessment report, NOT in any label.

A MEK-INHIBITOR-CONTAINING COHORT CANNOT SERVE AS A PARADOX READOUT AT ALL. The MEK inhibitor truncates the pathway one node downstream: it suppresses the paradox OUTPUT while leaving the MECHANISM untouched — proven by RAS-associated non-cutaneous malignancies still occurring on dabrafenib+trametinib at label tier. This invalidates claturafenib's entire ~60-patient ASCO 2026 cohort as paradox evidence and every dabrafenib class II clinical datapoint.

AN ANTIPROLIFERATIVE OR XENOGRAFT GROWTH SCREEN IS STRUCTURALLY INCAPABLE OF DETECTING PARADOXICAL ACTIVATION. Encorafenib's EU SmPC sentence 'does not inhibit RAF/MEK/ERK signalling in cells expressing wild-type BRAF' derives from exactly such a screen and is about failure to INHIBIT, not failure to ACTIVATE. This is the single most likely source of a mistaken 'encorafenib is paradox-sparing' inference, and the same design flaw underlies the Stuart AACR 2012 abstract on which four unretrievable claims depend.

NOT ONE PARADOX-BREAKER CLAIMANT HAS ANY HUMAN PHARMACODYNAMIC EVIDENCE OF THE MECHANISM. Mosperafenib's clinical pERK is entirely PK-DERIVED — no skin biopsy, no paired tumour biopsy, no normal-tissue pERK anywhere, across 80 patients. NXP200 has no clinical PD at any dose despite PD being a stated secondary objective. Claturafenib has none, on a monotherapy denominator of 18. Tovorafenib has tumour pERK but no skin PD, and its approved weekly schedule produced only 12%/8% median tumour pERK reduction at day 21 — so low target occupancy is an untested alternative explanation for its low cutaneous event rate.

THE 'NO cuSCC' PROOF-POINT IS FRAGILE WHEREVER IT IS USED. For mosperafenib it survives only in CONFERENCE abstracts and is absent from the peer-reviewed n=80 paper, in which BOTH dose-limiting toxicities were cutaneous. For claturafenib and NXP200 it is absence of REPORT, not a counted zero — no dermatologic surveillance protocol, no exposure duration, no denominator-time. For tovorafenib the population is paediatric (minimal UV-damaged skin) and NF1 patients were EXCLUDED — precisely the RAS-activated population where the preclinical signal sits.

THE ADELMANN PARADOX INDEX IS THE ONLY CROSS-AGENT NUMBER IN THE FIELD AND IT CANNOT BEAR THE WEIGHT PUT ON IT: a three-point correlation against pooled cross-trial rates with no recoverable denominator, one engineered keratinocyte line, no regulatory or cross-lab validation, and silent on dimer promotion (where Karoulia ranks dabrafenib WORSE than vemurafenib). Encorafenib's 3.7% figure from it is roughly HALF the label's randomised 8% and omits the separate 5% new-primary-melanoma signal. Tovorafenib, mosperafenib, NXP200 and claturafenib have NEVER been run through it, so none can be placed on the scale at all.

PARADOXICAL ACTIVATION IS NOT UNIFORMLY A LIABILITY IN THE PUBLISHED RECORD. Koya 2012 (peer, in vivo, vemurafenib itself) reports that 'vemurafenib paradoxically increased MAPK signaling, in vivo cytotoxic activity, and intratumoral cytokine secretion by adoptively transferred cells'; Ciccolini & Milano attribute reduced cetuximab skin toxicity in CRC to encorafenib's paradoxical ERK activation. Any differentiation argument that treats paradox as monotonically bad should anticipate this counter-framing.

SEVERAL COMPETITORS' OWN CORPORATE MATERIAL IS QUIETER THAN THEIR SCIENTIFIC LITERATURE — the opposite of the usual CI pattern. Roche's patient-facing page calls mosperafenib only 'an experimental medicine' with no mechanism claim; Pfizer's BRAFTOVI promotional MOA pages carry the paradoxical-activation statement and make NO paradox-breaker or dimer-selective claim; Day One's approved-product boilerplate makes no paradox or monomer/dimer claim; Array's 2015 differentiation pitch was tolerability, not paradox. The overstatement to attack is in posters, protocol introductions and review articles, not in the marketing.

'PARADOX BREAKER' AS A CLASS LABEL FOR CLATURAFENIB IS AN INTERNAL FORE ATTRIBUTION WITH NO PUBLIC VENUE. No public source in any tier applies the term to that molecule; the sponsor's own descriptors are 'next-generation BRAF dimer inhibitor' and 'pan-mutant BRAF inhibitor'. If the class label is used downstream it must carry 'internal Fore attribution, unmirrored in the public record' — otherwise Fore is the source of a claim it is auditing others for making.

TOVORAFENIB IS THE ONLY MOLECULE IN THE SET WITH A PUBLISHED DRUG-OCCUPIED RAF DIMER CO-CRYSTAL (6V34, 8F7O, 8F7P — both protomers occupied) AND A LABEL-STATED BINDING MODE IN BOTH JURISDICTIONS. Three of the six molecules have NO co-crystal at all (encorafenib, mosperafenib, NXP200); one has an unrefereed single-chain monomer deposit whose primary citation is still 'To be published' (claturafenib, 9BP8); vemurafenib's 3OG7 is an engineered AKAP9-BRAF fusion construct with 15-plus solubilising mutations. Structural evidence quality varies enormously across this set and is routinely overstated.

MOSPERAFENIB'S PARADOX-BREAKER DESIGNATION RESTS ON ONE UNREADABLE 2022 PAPER (Wichmann, Clin Cancer Res) PROPAGATED INTO THREE PEER-REVIEWED VENUES AND THREE CONGRESS ABSTRACTS WITHOUT NEW MECHANISTIC EVIDENCE. Naive evidence-counting scores three peer-reviewed 'paradox breaker' papers when only one contains a relevant experiment — and one of those three does not contain the string 'pERK' anywhere. Propagation count is not evidence weight, and this is the clearest worked example of the difference.

02Synthesis

SYNTHESIS — WHERE THE CLASS'S PARADOX CLAIMS ARE EVIDENCED, AND WHERE THEY ARE ASSERTION

Data cut: 1 September 2026.

1. The paradox itself is the best-evidenced proposition in the field — for the early-generation agents

For vemurafenib, dabrafenib and encorafenib the paradox is not a hypothesis. It is stated at label tier in both US and EU jurisdictions, operationalised into prescribing restrictions (a wild-type-BRAF non-indication for two of the three, a permanent-discontinuation rule for RAS-mutant non-cutaneous malignancies for all three), and backed by a randomised incidence difference (vemurafenib cuSCC/KA 24% vs <1%), molecular genotyping of the induced lesions (60%, 21/35, RAS-mutant, mostly HRAS Q61L), a clinical dechallenge/rechallenge case with an ex vivo mechanism, and an in vivo acceleration-and-MEK-blockade experiment.

The single cleanest fact in the whole audit needs no structure, no docking, no review and no inference: encorafenib's own label, in both jurisdictions, states BRAF V600E 0.35 nM, wild-type BRAF 0.47 nM, CRAF 0.30 nM — CRAF is inhibited more potently than the mutant target. Vemurafenib's SmPC Table 6 gives V600E 10 nM vs wild-type BRAF 39 nM (~4-fold), and its USPI says the molecule inhibits CRAF, ARAF and wild-type BRAF "at similar concentrations". Dabrafenib's window is ~4.9×–7.7×. The biochemical substrate for paradoxical activation is disclosed by the regulators themselves for all three early-generation agents.

2. Every "paradox breaker" claim in this set is weaker than the word implies

No molecule audited here can credibly claim paradox breaking in the absolute sense on the public record. Three distinct failure modes recur:

Mosperafenib — refuted by the sponsor's own readable data. The absolute claims ("does not trigger", "at all levels", "did not induce") are contradicted by Supplementary Figure S2 of the sponsor's own 2026 paper: pERK rises up to ~3.65× DMSO in a BRAF wild-type, KRAS-mutant line at concentrations inside the compound's own therapeutic window, and mosperafenib is not uniformly below encorafenib (it is the higher bar in SW480 at 1000 nM in both blocks). The designation is carried by one unreadable 2022 preclinical paper, propagated into three peer-reviewed venues and three congress abstracts without new mechanistic evidence. The sponsor's patent language is more honest than its journal language — but that patent expressly disclaims this molecule in claim 1, so it is not the sponsor's claim about mosperafenib at all. The most-quoted clinical proof-point (absence of cuSCC) survives only in conference material and is absent from the peer-reviewed report at n=80, in which both dose-limiting toxicities were cutaneous. The asset was removed from Roche's phase I pipeline as of 23 July 2026, for reasons Roche does not state.
NXP200 — one narrow experiment carrying an escalating verb. The entire mechanistic base is a single HCT116 (BRAF-WT / KRAS G13D) dimerisation observation with the assay not named, no dose-response, no quantification, no second RAS-mutant line, no keratinocyte or skin model and no in vivo paradox readout, against one comparator (dabrafenib). There is no label-tier and no peer-tier evidence for this molecule anywhere. And the sponsor's own peer-facing language is "significantly less paradoxical activation" and "spares wt-BRAF-containing RAF dimers" — relative, comparative, unquantified — which becomes "circumvents" and "physically disrupting" in company-tier material. The gap between those two registers is the finding, and no public experiment closes it.
Claturafenib — the claim is explicitly relative and concedes residual paradox. The peer-tier sentence is "less paradoxical activation of pERK than encorafenib" — it affirms residual paradoxical activation rather than its absence, and the magnitude is nowhere reported. The wild-type dimer co-IP is an over-expression experiment run in a non-matched system against an endogenous mutant arm. No public source in any tier calls this molecule a "paradox breaker" — the sponsor's own terms are "next-generation BRAF dimer inhibitor" and "pan-mutant BRAF inhibitor." The defining context for paradoxical activation — RAS-mutant, BRAF wild-type — has never been tested: no such cell line, no xenograft, no skin model, no patient cohort. And the entire public monotherapy safety denominator is 18 patients, with a silent skin section, no dermatologic surveillance protocol and unreported exposure duration.

3. Tovorafenib is the strongest case — and it is strong only when scoped

Tovorafenib has what none of the others have: a label-stated binding mode in both jurisdictions, three published co-crystals including a drug-occupied back-to-back dimer with both protomers bound, independent (non-sponsor) biochemistry and structural work, an EU label sentence explicitly covering "RAF monomers and dimers", and a prospectively pre-specified secondary-primary-malignancy AESI with a defined MedDRA search returning no SCC and no keratoacanthoma at N=137.

And yet: both approved labels carry an NF1 warning grounded in MAPK activation — USPI §13.2 verbatim, "increased phosphorylation of ERK at clinically relevant concentrations… suggesting activation, rather than inhibition, of the MAP kinase pathway." The sponsor's own 2025 paper shows a bell-shaped pERK curve in NF1-LOF lines. The originating primary source (Sun 2017) is dose-conditional in its own text ("Both MLN2480 and AZD628 give some paradoxical activation at subtherapeutic doses (0.01–0.1 µM)"), and reviews have stripped that qualifier. CHMP, on the sponsor's own data, wrote that "paradoxical activation could be an issue in this setting."

The honest formulation: tovorafenib's anti-paradox claim is true as scoped to BRAF fusions and false as generalised to RAS-pathway-activated, BRAF wild-type settings. That is not a small caveat — it is the exact setting in which paradox matters.

4. What separates a supported claim from an asserted one, in this record specifically

Five discriminators, each of which fires repeatedly here:

Was the molecule itself tested, or a tool compound / a class? A large share of the "vemurafenib paradox" literature is PLX4720 tool-compound literature — Hatzivassiliou, Heidorn, Sievert, and Poulikakos's definitive chemical-genetic panels all sit on the tool-compound side of that line, and the substitution is elided downstream. Dabrafenib has never been tested in any BRAF-fusion model. Both labels' §5.2 paradox sentences are class statements ("BRAF inhibitors"), not molecule-specific findings. A class statement in a label is a supported label claim, not a molecule-level experiment.
Does the assay design permit the observation the claim rests on? Encorafenib's "does not inhibit wild-type BRAF signalling" (EU SmPC §5.1) derives from antiproliferative and xenograft growth screens — structurally incapable of detecting paradoxical pERK induction. Absence of antiproliferative effect is not absence of paradox. Likewise, a MEK-inhibitor-containing cohort (claturafenib's ASCO 2026 cohort, every dabrafenib class II datapoint) cannot serve as a paradox readout at all, because the MEK inhibitor truncates the pathway one node downstream: it suppresses the paradox output while leaving the mechanism untouched.
Does the verdict axis get conflated with the narrative axis? An absence can be supported as a finding while contradicting a narrative. Encorafenib's EU label genuinely contains no paradox statement — that is a supported absence, not a "contradicted" claim, and its consequence is operational: anyone citing "the label" on encorafenib and paradox must specify WHICH label, because citing the EU SmPC alone yields a false negative.
Is the negative a search result or a source? "No source makes claim X" is an audit/search record, never peer or company tier, and can never be proven exhaustive — paywalled congress archives and internal decks were inaccessible throughout. Three audits had tagged such negatives as peer or company.
Does the claim harden as it travels? This is the most reliable tell in the entire dataset. Holderfield 2014's "It has been speculated that…" becomes a flat assertion downstream. Sun 2017's dose qualifier vanishes in the review literature ("without the paradoxical activation of the MAPK pathway"). Mosperafenib's ">25% reduction versus dabrafenib" patent language becomes "at all levels." NXP200's "did not induce dimers" becomes "physically disrupting the formation of the BRAF dimer." Claturafenib's "less paradoxical activation" becomes a blog's "significantly sparing normal cells." In every case the direction of travel is from a relative, conditioned, comparator-anchored primary statement to an absolute, unconditioned secondary one — and in every case the secondary statement is the one that reaches a competitive deck.

5. The quantitative comparison everyone reaches for is weaker than it looks

The Adelmann 2016 paradox index (vemurafenib 5.5 / dabrafenib 10 / encorafenib 50) is the only cross-agent number in the field, and it is: a three-point correlation against pooled cross-trial cuSCC rates aggregated from a multi-reference block with no recoverable denominator; run in one engineered keratinocyte line standing in for all paradox-competent tissue; and silent on dimer promotion, where Karoulia ranks dabrafenib worse than vemurafenib. It has no regulatory or cross-lab validation. Encorafenib's own 3.7% figure in that paper is roughly half the 8% cuSCC/KA the randomised COLUMBUS monotherapy arm later fixed in the label — and omits the separate 5% new primary melanoma signal entirely. Neither tovorafenib nor mosperafenib nor NXP200 nor claturafenib has ever been run through this assay, so none can be placed on the comparative scale at all.

6. Bottom line for a differentiation argument

Nothing in this competitive set has established paradox breaking in absolute terms on the public record. Two of the four claimants concede residual activation in their own peer-facing language; one is contradicted by its own supplementary data; one rests on a single unnamed-assay experiment.
The strongest evidenced position available to any molecule here is "attenuated, in a defined context, at a defined tier" — tovorafenib for BRAF fusions (label + peer + regulatory-assessment), claturafenib relatively versus encorafenib in a 19-line panel (peer, single paper, sponsor-generated, one hour, no RAS-mutant model).
The clinical proof-point everyone wants — a controlled cutaneous-neoplasia comparison — does not exist for any of the paradox-breaker claimants. Every one is single-arm, small, short, and reported without a dermatologic surveillance protocol. Where a comparator does exist it is confounded (COMBI-v: different BRAF inhibitor; every claturafenib and dabrafenib class II cohort: concurrent MEK inhibitor).
The one clean randomised design in the entire record is COMBI-d (n=423, double-blind, placebo-controlled, identical dabrafenib dose in both arms): cuSCC 2% vs 9%, hyperkeratosis 3% vs 32%. It measures MEK-inhibitor masking, not paradox breaking — and it does not abolish the RAS-driven tumour-promotion signal, which persists at label tier on the combination.
No human pharmacodynamic evidence of paradox absence exists for any claimant. Mosperafenib's clinical pERK is entirely PK-derived — no skin biopsy, no paired tumour biopsy, no normal-tissue pERK anywhere. NXP200 has no clinical PD at any dose despite PD being a stated secondary objective. Claturafenib has none. Tovorafenib has tumour pERK in phase 1 but no skin pharmacodynamics, and its approved once-weekly schedule produced only 12% / 8% median tumour pERK reduction at day 21 — so low target occupancy is an untested alternative explanation for its low cutaneous event rate.

03Claim-evidence matrix

CROSS-MOLECULE PARADOX-CLAIM AUDIT — CLAIM/EVIDENCE MATRIX

Data cut: 1 September 2026. Venue tiers: label (approved FDA USPI / EMA SmPC, section quoted) and peer are GROUNDING; regulatory-assessment (regulator-authored, NOT label) is grounding-quality but a distinct tier; conference is PROVISIONAL; company, review, aggregator/blog, vendor, audit/search-record and internal attribution are AWARENESS ONLY and can never on their own establish that a molecule does or does not cause paradoxical activation.

Six molecules are in scope: vemurafenib, dabrafenib, encorafenib (early-generation type I/I½), mosperafenib, NXP200, claturafenib (claimed paradox breakers), tovorafenib (type II / pan-RAF).

CLASS A — EARLY-GENERATION TYPE I / TYPE I½ (αC-OUT)

A1. Vemurafenib (ZELBORAF) — Genentech / Roche Registration GmbH

#	Claim	Who	Venue	Evidence basis	Verdict
V1	"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells…"	Genentech	label — USPI §5.2, rev. 07/2026	Label cites no study; underlying experiments not identified in the label	supported (as label text)
V2	"Based on mechanism of action, ZELBORAF may promote malignancies associated with activation of RAS…"	Genentech	label — USPI §5.1	Regulator/sponsor mechanistic inference, stated as such	supported
V3	"Based on mechanism of action, vemurafenib may cause progression of cancers associated with RAS mutations…"	Roche Registration GmbH	label — SmPC §4.4	Same; EU label never uses the word "paradox"	supported
V4	cuSCC/KA 24% vs <1% dacarbazine (336 vs 287); new primary melanoma 2.1% (7/336) vs 0; ECD cuSCC/KA 40.9% (9/22)	Genentech	label — USPI §5.1, Trials 1 and 4	Randomised incidence difference; USPI names no trial acronym	supported
V5	Myeloid neoplasms in ECD patients incl. those who received ZELBORAF; monitor CBC	Genentech	label — USPI §5.1	Label statement, no rate	supported
V6	CMML (NRAS-mut) and pancreatic adenocarcinoma (KRAS-mut) progression, frequency "Rare"	Roche	label — SmPC §4.8 tabulated ADRs, footnotes (4),(5); USPI §6.2	Frequency-categorised ADRs, not footnotes only	supported
V7	MEK co-inhibition reduces cutaneous events: cuSCC/KA 6% vs 20%; 2nd primary melanoma 0.8% vs 2.4%; BCC moves the other way 4.5% vs 2.4%	Genentech	label — COTELLIC USPI §5.1 (coBRIM)	Label states incidences, asserts no mechanism; mechanism strand added by Su 2012 in vivo MEK blockade (PLX4720)	partially-supported
V8	Non-cutaneous malignancies 0.8% (combo) vs 1.2% (vemurafenib arm)	Genentech	label — COTELLIC USPI §5.1	Label incidence, no RAS genotype confirmation	supported
V9	ATP-competitive RAF inhibitors including PLX4032 transactivate RAF dimers; drug acts in trans on the unbound protomer; requires RAS, CRAF, intact dimer interface	Poulikakos 2010 (Bollag = Plexxikon co-author)	peer — Nature 464:427–430, PMID 20179705	PLX4032 supplied purified catalytic-domain potency (35/110/48 nM) and pERK induction 40 nM–1 µM; the definitive chemical-genetic panels (gatekeeper mutant, kinase-dead V5-catC, R401 dimer-interface mutant) used PLX4720 and the JAB34 tool, not PLX4032	partially-supported (downgraded)
V10	PLX4032 activates ERK via RAF1 in BRAF-WT melanoma cells; enhances NRAS-mutant proliferation	Halaban 2010	peer — PMID 20149136	Vemurafenib itself tested; single most directly on-molecule preclinical paradox demonstration	supported
V11	RAF inhibitors prime wild-type RAF to activate MAPK and enhance growth	Hatzivassiliou 2010 (Genentech)	peer — PMID 20130576	GDC-0879 and PLX4720, not vemurafenib. "Independent replication" credit WITHDRAWN — only a Registered Report (protocol) exists	partially-supported
V12	BRAF inhibitors drive RAS-dependent BRAF–CRAF binding, CRAF activation, MEK-ERK signalling	Heidorn 2010	peer — PMID 20141835	PLX4720 and SB590885. Replication credit likewise withdrawn (protocol only)	partially-supported
V13	60% (21/35) of vemurafenib-associated cuSCC/KA carry RAS mutations, most prevalent HRAS Q61L	Su 2012 (first author Hoffmann-La Roche; Plexxikon co-authors; Roche-funded)	peer — NEJM 366:207–215	Discovery set 13/21 RAS of which 12 HRAS; validation 8/14. Corroborated by Oberholzer 2012	supported
V14	"…PLX4720 was not an initiator or a promoter of carcinogenesis but accelerated growth of the lesions harboring HRAS mutations, and this growth was blocked by concomitant treatment with a MEK inhibitor"	Su 2012	peer	In vivo HRAS Q61L skin carcinogenesis model. Carry-over caveat: PLX4720, not vemurafenib	supported (for PLX4720)
V15	Vemurafenib caused accelerated progression of pre-existing NRAS-mutant CMML; ERK hyperactivation ex vivo reversed on withdrawal	Callahan 2012	peer — NEJM 367:2316–2321, PMID 23134356	n=1 with multiple dechallenge/rechallenge cycles plus ex vivo mechanism — highest causal-inference clinical evidence in the whole record	supported
V16	">30-fold higher concentrations of vemurafenib were required to inhibit ERK driven by p61 V600E dimers than by p61 V600E R509H monomers"; pERK induced at low concentrations in WT-RAF/NRAS-mutant SK-MEL-30	Yao 2015	peer — Cancer Cell 28:370–383	Cellular, transfected SKBR3. Ratio only — the nanomolar values quoted downstream were invented and are struck	supported (ratio only)
V17	KIAA1549-BRAF homodimer resistant with paradoxical activation; paper treats vemurafenib paradox as "the well-described paradoxical activation"	Sievert 2013	peer — PNAS 110:5957–5962	Vemurafenib was never tested experimentally in this paper; compound was PLX4720. Downstream quotation attributed to it is not verbatim	asserted-only (for vemurafenib-relevant content)
V18	Vemurafenib paradox index 5.5 (EC80 ~2,100 nM ÷ A375 IC80 ~380 nM); peak pERK induction 6.86 ± 1.27×; tracks rank-order with clinical cuSCC 22%/6%/3.7%	Adelmann 2016	peer — Oncotarget 7:30453	HaCaT-HRAS^G12V; three-agent rank-order across separate trials; authors claim correlation not prediction	partially-supported
V19	Vemurafenib stabilises αC-OUT; reduced drug affinity for the second protomer (steric hindrance)	Karoulia 2016	peer — Cancer Cell 30:485–498, PMC5021590	Co-crystal geometry. Negative cooperativity is a structural inference, not a measurement. ~~R506 salt-bridge disruption~~ STRUCK — that formulation is a review artefact (Cope 2018), not in the primary	partially-supported
V20	Vemurafenib-induced paradoxical MAPK activation has a beneficial immunological consequence (T-cell function)	Koya 2012 (peer) + AACR 2012 Abstract 3510 (conference)	peer + conference	"vemurafenib paradoxically increased MAPK signaling, in vivo cytotoxic activity, and intratumoral cytokine secretion by adoptively transferred cells" — vemurafenib itself, in vivo	supported
V21	Canonical review framing of the RAF-inhibitor paradox	Holderfield 2014 NRC; Cope 2018; Lito/Solit	review	No new experiment. Cope 2018 is the origin of the R506 salt-bridge error	awareness only — cannot establish or refute
V22	No sponsor claim that vemurafenib avoids, breaks or mitigates paradoxical activation exists in the public record	Roche / Genentech / Plexxikon	audit/search-record (NOT company)	A negative search result cannot carry a truth verdict	`no-claim-located`
V23	Implicit label claim by omission: MOA section describes vemurafenib purely as a kinase inhibitor — "dimer" appears nowhere in the USPI, "paradox" zero times in the entire SmPC	Genentech USPI §12.1; SmPC §5.1	label	Grep-verified absences. Qualified: USPI §1.1 Limitation of Use cross-references §5.2 to a prescribing restriction	asserted-only, partly qualified

Label-tier biochemical substrate (SmPC §5.1 Table 6): BRAF V600E 10 nM vs BRAF wild-type 39 nM — only ~4-fold. USPI §12.1: "Vemurafenib also inhibits other kinases in vitro such as CRAF, ARAF, wild-type BRAF… at similar concentrations." The precondition for the paradox is disclosed in the label itself.

A2. Dabrafenib (TAFINLAR) — Novartis (originator GSK)

#	Claim	Who	Venue	Evidence basis	Verdict
D1	"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells which are exposed to BRAF inhibitors. Confirm evidence of BRAF V600E or V600K mutation status prior to initiation…"	Novartis	label — USPI §5.2, rev. 3/2026	Label cites no study, cell line or concentration, and states the finding at class level ("BRAF inhibitors"), not for dabrafenib. The dabrafenib-specific experiment (King 2013, GSK-authored) is not referenced	supported (as label text)
D2	"TAFINLAR is not indicated for treatment of patients with wild-type BRAF solid tumors [see Warnings and Precautions (5.2)]"	Novartis	label — USPI §1.8 Limitations of Use	A paradox-driven negative indication cross-referenced to the tumour-promotion warning	supported
D3	"Based on its mechanism of action, TAFINLAR may promote the growth and development of malignancies with activation of RAS…"; §2.4 Table 5 "Non-Cutaneous RAS Mutation-positive Malignancies — Permanently discontinue TAFINLAR"	Novartis	label — USPI §5.1 + §2.4	Label's own wording is explicit mechanistic inference; the 1% non-cutaneous malignancy incidence is not RAS-genotype-resolved, so rule and number are not linked by evidence	partially-supported
D4	RAS-associated non-cutaneous malignancies on dabrafenib monotherapy (pancreatic adenocarcinoma, bile duct adenocarcinoma) and on dabrafenib+trametinib (colorectal, pancreatic); CMML and non-cutaneous head-and-neck SCC attributed to "another BRAF inhibitor"	Novartis	label — SmPC §4.4	Strongest dabrafenib-specific clinical paradox evidence, at label tier. Qualified: SmPC does not state RAS mutation was confirmed in the individual cases	supported, qualified
D5	"HCT-116 cells showed increased pMEK and pERK following dabrafenib treatment at 100 nM or 300 nM"; "MAPK upregulation was not sensitive to ARAF or BRAF knockdown, but was reduced to baseline upon CRAF depletion"	King 2013 (GSK employees)	peer — PLOS ONE 8:e67583	siRNA epistasis — CRAF dependence directly demonstrated. Paper does not itself measure RAF dimerisation	supported
D6	MEK inhibitor abrogates the paradox in vivo (skin)	King 2013	peer	In-vivo skin arm used a 12-day rat study with tool compounds GSK2366297 + GSK2091976, not dabrafenib + trametinib. Cellular pERK-abrogation arm is the better-grounded half	partially-supported
D7	cuSCC 7% (dab+tram 150/2) vs 19% (dab mono), P=0.09	Flaherty 2012, BRF113220	peer	Open-label; 247 enrolled, 85 phase 1, 162 randomised across three arms; authors' own conclusion "nonsignificantly reduced". Widely quoted without P=0.09	partially-supported
D8	"fewer patients in the dabrafenib–trametinib group than in the dabrafenib-only group had incident cutaneous squamous-cell carcinoma (2% vs. 9%) or cutaneous hyperkeratoses (3% vs. 32%)"	Long 2014, COMBI-d	peer	n=423, double-blind, placebo-controlled, identical dabrafenib dose in both arms — the only design where the MEK inhibitor is the sole variable	supported
D9	cuSCC/KA 1% (combination) vs 18% (vemurafenib)	Robert 2015, COMBI-v	peer (retagged from conference)	n=704, open-label; confounded — comparator is a different BRAF inhibitor as monotherapy, so the delta conflates MEKi masking with an inter-agent difference	partially-supported
D10	"PB, DAB and VEM showed potent inhibition of monomeric BRAF^V600E but 30 to 100-fold less potency for inhibition of dimeric BRAF^V600E"	Karoulia 2016	peer	Paired biochemical/cellular monomer vs enforced-dimer assays. The dimer potency loss is shared across the compared αC-OUT compounds — it is not dabrafenib-discriminating	supported
D11	"DAB and AZ-VEM, although αC-OUT inhibitors, promote RAF priming and dimerization more potently than VEM and PB and at levels closer to αC-IN inhibitors"; R506 held in a further IN position	Karoulia 2016	peer	Crystallographic + priming/RAS-GTP complex assays. Counter-intuitive and verified. AZ-VEM is a chimeric tool compound, not vemurafenib and not a clinical asset	supported
D12	"It has been speculated that the lower incidence of cutaneous lesions seen with dabrafenib is owing to higher potency against BRAF^V600E compared with WT BRAF and CRAF, whereas vemurafenib may be relatively equipotent"	Holderfield, Nagel & Stuart 2014 (Holderfield = UCSF; Nagel and Stuart = Novartis NIBR)	review	No new experiment; the source itself marks this as speculation. Contradicted in tension by the label's modest ~4.9×/7.7× selectivity window and by D11	asserted-only
D13	Dabrafenib paradox index 10; A375 IC80 5.9 nM; HaCaT-HRAS^G12V pERK EC80 59 nM; peak pERK induction 2.76 ± 0.34×; clinical cuSCC ~6%	Adelmann 2016	peer	Three-point correlation; pooled cross-trial rates; one engineered keratinocyte line; index silent on dimer promotion, where Karoulia ranks dabrafenib worse than vemurafenib	partially-supported
D14	Dabrafenib paradoxically activates BRAF-fusion-driven tumours	(class inference)	—	Dabrafenib has never been tested in any BRAF-fusion model. Sievert 2013 used PLX4720 and a second-generation Plexxikon compound; Jain 2017 also did not test dabrafenib	asserted-only / unmeasured class inference
D15	"no label, peer, conference or company source anywhere makes a dimer-selectivity or paradox-breaking claim for dabrafenib"	audit's own search	audit/search-record (peer tag withdrawn)	A negative search result is not a source	usable only as "none located in this pass"
D16	GSK approval press release restating §5.2 verbatim + BREAK-3 figures (cuSCC 7%; new primary melanoma 2% (3/187) vs none on chemotherapy)	GSK, 29 May 2013	company	Label-restatement material. Company tier is populated, not empty — but establishes nothing	awareness only

Label-tier biochemistry (USPI §12.1): BRAF V600E 0.65 nM; wild-type BRAF 3.2 nM; CRAF 5.0 nM — ~4.9× to ~7.7× window. No assay format or oligomeric state disclosed. "dimer" and "monomer" appear ZERO times in the 57-page USPI and ZERO times in the 68-page SmPC.

A3. Encorafenib (BRAFTOVI) — Pfizer / Pierre Fabre

#	Claim	Who	Venue	Evidence basis	Verdict
E1	"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells, which are exposed to BRAF inhibitors. Confirm evidence of BRAF V600E or V600K mutation prior to initiating BRAFTOVI."	Pfizer	label — USPI §5.2, rev. 2/2026	Label cites no specific study; class statement applied to encorafenib. Encorafenib-specific grounding comes from outside the label	supported
E2	"Based on its mechanism of action, BRAFTOVI may promote malignancies associated with activation of RAS… Discontinue BRAFTOVI for RAS mutation-positive noncutaneous malignancies."	Pfizer / EMA	label — USPI §5.1; SmPC §4.4	Regulator-mandated restriction in both jurisdictions; no trial or experiment cited in either label	supported
E3	"Encorafenib does not inhibit RAF/MEK/ERK signalling in cells expressing wild-type BRAF"	Pierre Fabre	label — EU SmPC §5.1	Traceable to antiproliferative/xenograft growth screens (Stuart AACR 2012). A statement about failure to INHIBIT, not about failure to ACTIVATE. Contradicted in-document by the same §5.1's WT BRAF 0.47 nM / CRAF 0.30 nM and by §4.4's RAS discontinuation rule	partially-supported — the single most likely source of a mistaken "paradox-sparing" inference
E4	The EU Product Information contains no paradox statement and no equivalent to USPI §5.2	(absence, documented by reading the label)	label (absence)	"paradox", "paradoxical" and "dimer" do not appear anywhere in the retrieved UK emc SmPC (Annex I equivalent)	supported as an absence. Bearing: anyone citing "the label" on encorafenib and paradox must specify WHICH label
E5	"The encorafenib dissociation half-life was >30 hours and resulted in prolonged pERK inhibition"	Pierre Fabre	label — EU SmPC §5.1 (EU only; the FDA label omits it entirely)	Measured from V600E-mutant BRAF (Koelblinger 2018). Carries no information about dwell time on wild-type BRAF or CRAF — where paradox happens	supported; bearing neutral
E6	"Notably, Encorafenib pre-treatment promotes paradoxical activation in these cells" — panel includes H2087 (BRAF L597V, class II), H1666 (BRAF G466V, class III), SKMEL-30 and SKMEL-2 (BRAF-WT/NRAS Q61)	Cotto-Rios 2020 (Einstein/Mount Sinai)	peer — Nat Commun 11:4370	Direct experiment; pERK immunoblot with pre-treatment and washout. Figure 7 legend; tool condition 1 µM for 1 h — single high concentration, not a dose–response	supported, qualified
E7	Encorafenib paradox index 50 (A375 IC80 2.6 nM; HaCaT-HRAS^G12V pERK EC80 130 nM); peak pERK induction 4.08 ± 0.16×	Adelmann 2016	peer	A paradox index of 50 requires that a pERK-induction EC80 was measurable — the paper documents that encorafenib DOES paradoxically activate ERK and quantifies only the dose separation	supported
E8	The index predicts clinical toxicity: cuSCC ~22% / 6% / 3.7% for vemurafenib / dabrafenib / encorafenib	Adelmann 2016	peer	Ecological correlation, multi-reference block, no recoverable denominator. The 3.7% figure is ~half the label's randomised single-agent 8% and omits the 5% new-primary-melanoma signal	partially-supported
E9	Time-dependent MAPK reactivation with encorafenib in BRAF-V600E lines at 3/24/48 h	Pickles 2020	peer — Oncotarget 11:3188	Direct experiment (RealTime-Glo + pERK immunoblot)	supported
E10	"Group 1", "negative co-operativity", "paradoxical pathway activation in BRAF wild-type cells"	Pickles 2020	peer, but Introduction re-cited to refs [4]–[6]	The paper tested no BRAF wild-type line and no RAS-mutant line and performed no experiment measuring paradoxical activation	asserted-only
E11	"The experimental Type 1 RAF inhibitor, SB590885, and/or encorafenib… increased ERK1/2 phosphorylation in cardiomyocytes, and promoted hypertrophy, consistent with a 'RAF paradox' effect… in mouse hearts in vivo"	Biochem J 2022;479(3):401–424, PMID 35147166	peer, in vivo	Clearest demonstration that encorafenib's paradox is not confined to keratinocyte tissue culture	supported
E12	Human-tissue confirmation: eruptive naevi, pyogenic granulomas, second primary melanomas on encorafenib, two with molecular genotyping confirming BRAF wild-type status in the induced lesion	Multiple independent dermatology/oncology groups	peer (case level)	Strongest human-tissue evidence that the operating mechanism is paradoxical activation of the wild-type allele. No denominators, no control arm	supported (existence, not rate)
E13	"Fewer cetuximab-related skin-toxicities… a Yin and Yang effect of ERK paradoxical activation"	Ciccolini & Milano 2022	review/editorial	No new experiment, no denominators	asserted-only — but directionally incompatible with a paradox-sparing narrative
E14	"BRAF inhibition with encorafenib exhibits substantial antitumor activity with less paradoxical MAPK pathway activation leading to treatment resistance"	Carr, Sun, Eroglu, Zager 2020	review	No head-to-head paradox experiment cited; "less" unquantified; conflates paradox with acquired resistance	asserted-only — the archetypal sentence to watch for in a competitor deck
E15	The combination "shows peculiar pharmacodynamic properties which translate in a higher on-target potency and paradox index"	Indini & Mandalà 2020	review	Pure re-citation of Adelmann. A high paradox index is a dose-separation statistic, not an absence of paradox	asserted-only
E16	Gastric hyperplasia (mouse cuSCC surrogate) "does not even occur at the most efficacious dose of encorafenib"	Koelblinger, Thuerigen (employee, Pierre Fabre Pharma GmbH), Dummer 2018	review	Cited to Stuart AACR 2012, whose full text is confirmed non-retrievable (HTTP 403) and does not contain gastric hyperplasia in any indexed excerpt. A review paraphrasing an inaccessible conference abstract	asserted-only
E17	"LGX818 was inactive against BRAF wild-type tumors at doses up to 300 mg/kg bid"; regressions at ≥1 mg/kg in BRAF-mutant xenografts	Stuart 2012, AACR Abstract 3790	conference	Verified in indexed abstract text. The >400-cell-line screen, the >30 h off-rate, the gastric-hyperplasia surrogate and the A375 pERK/antiproliferative EC50s are NOT in the retrievable text and reach the record only via review paraphrase	asserted-only. A growth-inhibition screen is structurally incapable of detecting paradoxical pERK induction
E18	"wide therapeutic index"; differentiated safety	Array BioPharma PR, 1 June 2015	company	Traced to the non-retrievable Stuart abstract. NEGATIVE FINDING: the release makes no claim about paradoxical activation, cuSCC, KA, RAF dimers or paradox breaking — the sponsor's differentiation pitch was tolerability	asserted-only / awareness
E19	Pfizer HCP MOA pages carry the paradoxical-activation statement	Pfizer	company	NEGATIVE FINDING: no "paradox breaker", "dimer-selective", "equipotent monomer/dimer" or "no RAF transactivation" claim anywhere in the sponsor's own promotional material	awareness only
E20	No source in any venue applies "paradox breaker", "dimer-selective", "equipotent against monomer and dimer", "no RAF transactivation" or "does not induce paradoxical activation" to encorafenib	audit's own search	audit/search-record (peer tag withdrawn)	Contradicting primary evidence is affirmative, not merely absent (E6, E7, E11, E1)	supported as a negative search result; not provably exhaustive

The single cleanest label-tier fact in the whole audit: FDA §12.1 and EU §5.1, both jurisdictions verbatim — BRAF V600E 0.35 nM, wild-type BRAF 0.47 nM, CRAF 0.30 nM. Encorafenib has no enzymatic selectivity for the mutant target, and CRAF is inhibited more potently than BRAF V600E. That is the biochemical substrate for paradox, stated by both regulators, in the same paragraph that calls the molecule "potent and highly selective". FDA §12.1 further discloses binding to JNK1/2/3, LIMK1/2, MEK4 and STK36 at ≤0.9 µM.

Class-level correction that retires a whole line of argument: Tkacik 2026 (eLife RP110344) reports that type I AND type II inhibitors induce paradoxical activation of all three RAF isoforms, with the type II inhibition phase suggestive of positive cooperativity. Paradoxical activation is therefore not a property of the αC-OUT / type I½ class alone, and no molecule's binding class can be used to argue paradox in either direction. [FLAG: Tkacik does not test encorafenib — it names it once, in the Introduction only. It is used here solely to refute a class-exclusivity generalisation, never as evidence about encorafenib itself.]

CLASS B — CLAIMED PARADOX BREAKERS

B1. Mosperafenib (RG6344 / RO7276389) — F. Hoffmann-La Roche. LABEL TIER EMPTY (verified by retrieval).

#	Claim	Who	Venue	Evidence basis	Verdict
M1	"does not trigger pERK overactivation in BRAF wild-type contexts"	Renner, Wichmann, Pettazzoni et al. (Roche)	peer — Mol Cancer Ther 2026;25(4):599–609	The programme's own readable Suppl Fig S2 refutes it. Mosperafenib raises pERK above baseline in all three BRAF-WT lines, peaking at ~3.65× DMSO (HCT116, 100 nM + cetuximab) and ~2.85× in monotherapy, with the bell-shaped concentration–response signature of paradoxical activation	CONTRADICTED
M2	Substantially less paradoxical activation than encorafenib	Same	peer — Fig 3A-C / Suppl Fig S2	Direction holds at most concentrations; largest separation SW48 1000 nM (~1.6× vs ~3.65×). Does NOT hold universally — in SW480 at 1000 nM mosperafenib is the HIGHER bar in both the monotherapy and +cetuximab blocks. n unstated, no statistics, three CRC lines, no skin-relevant model	partially-supported — establishes attenuation, not abolition, and not uniform attenuation
M3	"proves paradox breaking properties by not triggering MAPK over-activation in BRAF WT contexts at all levels"	Renner et al.	conference — AACR 2025 Abstract 5606	No cell lines, concentrations or denominators given. The absolute quantifier is directly falsified by the same programme's later readable data	CONTRADICTED — the single most overstated formulation found anywhere
M4	"did not induce RAF paradoxical activation" (Compound Ia)	Wichmann 2022	peer — Clin Cancer Res 28:770–780	THE FOUNDATIONAL CLAIM — evidence could not be verified. Full text Cloudflare-blocked; abstract names no cell line, no concentration, no assay. Paper never uses the names mosperafenib / RG6344 / RO7276389	UNKNOWN
M5	"None of the typical BRAFi class toxicities… have been observed to date, highlighting the paradox breaking properties"	Fontana E et al.	conference — ASCO 2025 Abstract 3542, n=51	Factual half (zero cSCC, PPE, KA) is a real single-arm observation. Inferential half is an unsupported causal leap — no on-treatment skin biopsy, no tissue pERK, no comparator arm	ASSERTED-ONLY
M6	Same absence statement without the causal clause; grade 3 TRAEs 11/66; one DLT of grade 3 rash maculo-papular	Vieito M et al.	conference — AACR 2025 CT017, n=66	Descriptive observation stands as fact; mechanism not established by it; title's "paradox breaker" is imported	ASSERTED-ONLY as a paradox claim
M7	Title designation "paradox breaker"; "There were no reports of palmar-plantar erythrodysesthesia or keratoacanthoma"	Vieito, Fontana et al.	peer — J Clin Oncol 2026;44(14):1337–1348	MATERIAL FINDING: the claim shrank on the way to peer review. Both conference abstracts (n=51, n=66) asserted cuSCC had not been observed; the peer-reviewed paper at n=80 does not mention cSCC. Two DLTs, both cutaneous (grade 3 rash; grade 3 rash maculopapular). No paradox experiment of its own; PD is PK-derived	ASSERTED-ONLY
M8	"Mosperafenib is a novel, paradox-breaking BRAF inhibitor"	Serrano-Serrano et al.	peer — Cancer Res Commun 2026;6(6):1435–1446	None. Full text read: the string "pERK" does not occur anywhere in the paper. No wild-type or RAS-mutant model, no paradox experiment	ASSERTED-ONLY — recorded because propagation count is itself a CI signal: naive evidence-counting scores three peer-reviewed "paradox breaker" papers when only one contains a relevant experiment
M9	Compounds of the claimed formula "show considerably less paradoxial [sic] activation of the MAPK signalling pathway while retaining high potency"	F. Hoffmann-La Roche, WO2021116050A1 "New BRAF inhibitors as paradox breakers"	company (patent)	CLAIM 1 EXPRESSLY EXCLUDES MOSPERAFENIB by proviso, repeated in claims 1, 2, 3 and throughout the embodiments. Assay is HTRF pERK A375 vs HCT116, comparator dabrafenib (+ AR-25); quantified conclusion is "the maximum paradox inducing effect is substantially reduced by more than 25% for all examples" — i.e. every exemplified compound still induces a measurable paradox signal	UNKNOWN as to mosperafenib — the patent is not about this molecule
M10	Classifies "Compound Ia" among paradox breakers; conformation NA; cell-free IC50 BRAF(WT) <5E-10, BRAF V600E 1.77E-09; dimer column ("confirmed experimentally") = BRAF/BRAF	Acta Pharm Sin B 2024	review	Secondary throughout; generated no data; cites Bonfill-Teixidor 2022 and Wichmann 2022 (both paywalled)	UNKNOWN — awareness only
M11	"'paradox-breaker' BRAF inhibitors (e.g., PLX8394, PLX7904, mosperafenib) inhibit RAF dimerization and maintain MAPK suppression"	Biomolecules 2026;16:543 (independent authors)	review	None. The only source located that makes an explicit RAF-dimerisation claim about mosperafenib by name, grouping it with others by assumption. A live contamination route into CI files and LLM summaries	ASSERTED-ONLY
M12	Roche's patient-facing trial page describes RO7276389 only as "an experimental medicine" — no paradox, paradox-breaker, RAF-dimer or mechanism claim of any kind	Roche	company	Null result, informative. The overstatement here occurs in the scientific literature, not in the marketing	UNKNOWN (informative null)
M13	"potent, selective, brain penetrant BRAF inhibitor with binding Kd of 0.6, 1.2 and ~1.7 nM for BRAF WT, BRAF V600E and c-RAF, presenting paradox breaker properties"	ProbeChem / MedKoo / Selleck / MCE	company (vendor)	ProbeChem does cite Wichmann 2022 as its primary source; the cited primary is unreadable. Pattern corroborated by the Acta review's table	ASSERTED-ONLY. Both attestations describe a compound binding wild-type BRAF at least as tightly as V600E — a pan-RAF-like binding profile nobody has reconciled with paradox-breaker positioning
M14	Development status	Roche Group development pipeline, pharma half-year 2026, status 23 July 2026, p.2	company	"RG6344 mosperafenib (BRAFi) – solid tumors" is listed under "Removed from phase I." Roche gives no reason — do not infer that paradox behaviour, cutaneous toxicity or efficacy caused it	supported (fact of removal only)
M15	New finding not in the sponsor's paradox framing: under mosperafenib 300 nM in HT29, P-ERK rebounds at 24–48 h and P-c-RAF is induced, in both −EGF and +EGF arms; cetuximab blunts both	Suppl Fig S3, MCT 2026	peer (supplement)	This is EGFR-feedback reactivation in a V600E line, not classical wild-type dimer paradox — do not conflate. But it is an on-pathway RAF-activation readout for this molecule	supported (read from blot; see flag)

B2. NXP200 (HSK42360 / HSK42360-Na) — Haisco / Nuvectis. LABEL TIER EMPTY and PEER TIER EMPTY (both verified by retrieval).

⚠️ Three of the substantive sources carry journal names and journal DOIs (Cancer Research, J Clin Oncol, Neuro-Oncology) and are supplement-published conference abstracts, not peer-reviewed primary literature. Re-tagging 10.1200/jco.2025… or 10.1093/neuonc/noaf201… as "peer" because of the journal name commits precisely the venue-integrity error this audit exists to prevent.

#	Claim	Who	Venue	Evidence basis	Verdict
N1	"did not induce homodimer or heterodimer of CRAF with CRAF or BRAF and thereby blocked reactivation of the MAPK pathway in HCT116 cells, whereas Dabrafenib significantly induced RAF dimerization and MAPK reactivation"	Gao et al., Haisco	conference — AACR 2024 #598	This is the experiment — the only paradox-specific data in the entire record. One BRAF-WT / KRAS G13D line. Assay not named. No dose-response, no quantified blot, no second RAS-mutant line, no keratinocyte/skin model, no in vivo paradox readout. Comparator: dabrafenib only	partially-supported
N2	Title designation "a potent, brain permeable, BRAF paradox breaker"	Gao et al.	conference	Rests entirely on N1 plus a 5 nM enzymatic IC50 and an unquantified cellular WT-selectivity statement	partially-supported
N3	"displays significantly less paradoxical activation than approved BRAF inhibitors and spares wt-BRAF-containing RAF dimers"	Li J / Li W et al.	conference — ASCO 2025 #3109; SNO 2025 CTNI-28	Stated as background in both abstracts. No data shown for either half. "Spares wt-BRAF-containing dimers" is a structural-selectivity claim with zero structural or biochemical support	ASSERTED-ONLY. Note it is relative and unquantified — the honest version of the company's absolute language
N4	"excellent and durable anti-tumor effect in BRAF Class I and II mutant CDX or PDX models"	Li J / Li W et al.	conference	No class II mutant named, no model named, no TGI or volume data, no figure. Background sentence only	ASSERTED-ONLY
N5	"effectively inhibits signaling and disrupts BRAF homo- and hetero-dimers, with significantly reduced paradoxical activation"	Gao et al.	conference — AACR 2026 LB405	LB405 does present preclinical results (glioma PDX TGI 67/108/111.5%, Kp,uu 1.09 mouse / 2.8 rat, vemurafenib+cetuximab-resistant CRC model, trametinib synergy) — but no dimerisation method, assay, dose-response, quantification or figure accompanies the mechanistic sentence. "Disrupts" is a stronger verb than "failure to induce" supports	ASSERTED-ONLY
N6	"NXP200 circumvents this by physically disrupting the formation of the BRAF dimer, resulting in inhibition of the Class I mutation, and also of the dimerizing, RAS independent (Class II) and RAS-dependent (Class III) mutations"	Nuvectis Pharma	company — pipeline page, no references or footnotes on the page	Active physical disruption is not established by N1 (failure to induce ≠ disruption). No structural or biophysical data exist. The class III half has no supporting experiment of any kind anywhere in the record	ASSERTED-ONLY
N7	"paradox breaking properties… while reducing toxicities associated with their usage"	Nuvectis Pharma	company	No comparative toxicity data of any kind — no comparator arm, no cross-trial toxicity comparison with numbers, no cutaneous reporting	ASSERTED-ONLY
N8	"best-in-class BRAF paradoxical breaker inhibitor"	Haisco	company	No head-to-head against any other paradox breaker or type II RAF inhibitor is published	ASSERTED-ONLY
N9	Sodium salt: "early data indeed demonstrate a marked improved PK and greater single agent clinical activity"	Nuvectis Pharma	company	No PK parameters, exposure figures, response numbers, denominators or comparison shown	ASSERTED-ONLY
N10	">40% response rate in low- and high-grade adult glioma, including one Complete Response. Durable responses also in NSCLC, colorectal and papillary thyroid cancers"	Nuvectis Pharma	company	Glioma corroborated by LB405 (40.9%, 9/22, 1 CR) — though the HGG component 36.8% (7/19) explicitly "includ[es] preliminary PR". CRC weakly corroborated (1 PR / 11 evaluable at n=17). NSCLC and papillary thyroid: no patient numbers, denominators or durations exist anywhere. "Durable" has no supporting duration figure for any tumour type	split — glioma partially-supported; NSCLC / thyroid / "durable" asserted-only
N11	"a BRAF paradox breaker with efficacy in models of brain metastasis and BRAF-inhibitor resistance"	BioWorld (trade press); Synapse/PatSnap and Chinese aggregators	review (trade press / database aggregation)	No independent data; restates sponsor copy; BioWorld full text not retrievable (HTTP 403)	ASSERTED-ONLY — awareness only
N12	Cross-trial efficacy benchmark: "An approved BRAF V600E inhibitor combined with a MEK inhibitor, had ORRs of 33% and 50% in BRAF V600E-mutated HGG and LGG, respectively (dabrafenib US Prescribing Information, 2025)"	Li W et al.	conference invoking a label for a different molecule	Naive cross-trial comparison — different trial, different population, different era, no adjustment. The only label-tier document invoked anywhere in the NXP200 record, and it is not NXP200's label	awareness only

The escalation chain is the finding. Rung A (conference): "did not induce dimers" — the actual experiment. Rung B (conference): "significantly less paradoxical activation… spares wt-BRAF-containing dimers" — relative, unquantified, no data shown. Rung C (conference): "disrupts dimers, significantly reduced paradoxical activation". Rung D (company): "circumvents… physically disrupting" — absolute. No public experiment closes the gap between rung A and rung D, and no source quantifies the reduction.

B3. Claturafenib (PF-07799933 / ARRY-440) — Pfizer (originator Array BioPharma). LABEL TIER EMPTY (verified by retrieval — PubMed returns exactly two records for this molecule).

#	Claim	Who	Venue	Evidence basis	Verdict
C1	"In BRAF wild-type cells, PF-07799933 demonstrated no pERK inhibition, in contrast to the pan-RAF dimer inhibitors, and less paradoxical activation of pERK than encorafenib"	Yaeger R et al.; senior author Rothenberg SM (Pfizer)	peer — Cancer Discov 2024;14(9):1599–1611	Quantitative pERK in-cell Western dose–response, 19-line panel including BRAF wild-type lines, 1 h, four-parameter fit, head-to-head comparators (encorafenib, belvarafenib, tovorafenib, naporafenib, exarafenib). Wild-type IC50 ≥9,800 nmol/L. The claim is explicitly comparative and relative — less, not absent. No fold-induction value, induction EC50 or paradox index is reported in the text. Not a RAS-mutant paradox model, not in vivo, not a skin model	partially-supported
C2	"…disrupted endogenous mutant-BRAF:wild-type-CRAF dimers, and spared wild-type ERK signaling"	Yaeger, Rothenberg et al.	peer	Two co-IP experiments in non-matched systems — endogenous mutant arm (MEL21514, A375-NRAS^Q61K) vs HEK293T over-expression wild-type arm — plus ITDR showing 10-fold higher affinity for BRAF^V600E than wild-type CRAF. No endogenous wild-type dimer experiment; no functional RAS-driven wild-type dimer readout	partially-supported
C3	Cocrystal conformation explains paradox avoidance: "induced an ordered A-loop and caused an outward rotation in the αC-helix, both consistent with an inactive kinase conformation that may destabilize RAF dimers or prevent dimerization"	Yaeger, Rothenberg et al.	peer (the Cancer Discovery sentence only)	One X-ray cocrystal (BRAF-KDL, MR from 4MNE via PHASER). Hedged by the authors themselves. No biophysical dimer-affinity measurement (SPR, SEC-MALS, AUC, cross-linking). PDB 9BP8 is a monomer and an unrefereed deposit ("To be published") — not peer tier	partially-supported
C4	"Since it does not affect wild-type (WT) dimers, it has a favorable safety profile in the clinic"	Scotece & Drosten (independent commentators)	review — Trends in Cancer 2024;10(7):576–578	None of its own. The causal bridge is not established by the underlying paper; the clinical dataset is 30 phase 1 patients in whom the paradox-toxicity endpoint was never affirmatively measured or reported. Restates the wild-type dimer finding without the over-expression caveat	ASSERTED-ONLY
C5	"Works by disrupting the formation of BRAF-containing dimers" / "significantly sparing normal (wild-type) cells to reduce off-target side effects"	New Drug Approvals, self-published monograph blog	aggregator/blog (lowest awareness tier)	Cites no primary literature. Materially inverts the mechanism — the primary data show disruption of mutant-BRAF-containing dimers and failure to disrupt wild-type BRAF:CRAF dimers	CONTRADICTED
C6	"Next-generation BRAF dimer inhibitor"	Pfizer; presenter Monica Chen (MSK)	conference — ASCO 2026 Abstract 9512	A descriptor in a title, not a data-backed claim. The sponsor does not use "paradox breaker" and makes no paradox claim in its own conference or press material. All reported content is clinical; every patient received a concurrent MEK inhibitor	UNKNOWN as a paradox claim
C7	"Selective for inhibition of mutated BRAF, which may contribute to a favorable therapeutic safety profile"	Pfizer Oncology Development molecule page	company (with vendor catalogues at no tier and NCI Thesaurus at tertiary database)	Derivative restatement of the ≥9,800 nmol/L wild-type figure. No independent experiment. Safety inference forward-looking and tied to no reported cutaneous-toxicity dataset	ASSERTED-ONLY
C8	Classification as a "paradox breaker"	Fore CI registry	NO VENUE / INTERNAL ATTRIBUTION ONLY	No public source in any tier applies the term to this molecule. The primary paper uses "paradox"/"paradoxical" in exactly two sentences — one generic background statement and one relative claim that affirms residual paradoxical activation rather than its absence	ASSERTED-ONLY. If used downstream it must carry "internal Fore attribution, unmirrored in the public record"
C9	Absence-of-paradox-toxicity by silence	(no peer source makes this claim; the peer source is silent)	review / inference — venue of a claim is the venue of whoever makes it, not of the document that fails to mention it	ABSENCE OF REPORT, NOT EVIDENCE OF ABSENCE. Yaeger 2024's AE threshold was "TEAEs reported in ≥3 patients"; the safety section is simply silent on cuSCC / KA / new primary melanoma / skin papilloma / hyperkeratosis; no skin-surveillance protocol, dermatologic schedule or explicit negative statement is reported. The ASCO 2026 cohort (~60) is invalidated as a paradox readout because every patient received a concurrent MEK inhibitor. NCT05355701 (n=103) has posted no results	UNKNOWN

Public monotherapy safety denominator for this molecule is 18 patients, with a silent skin section, no dermatologic surveillance protocol and unreported exposure duration. Any assertion that claturafenib's clinical safety demonstrates paradox breaking is currently unfalsifiable, not validated.

CLASS C — TYPE II / PAN-RAF

C1. Tovorafenib (OJEMDA) — Day One Biopharmaceuticals / Ipsen Pharma. The only claimed paradox-sparing molecule with an approved label in this set.

#	Claim	Who	Venue	Evidence basis	Verdict
T1	"Tovorafenib is a Type II RAF kinase inhibitor of mutant BRAF V600E, wild-type BRAF, and wild-type CRAF kinases. Tovorafenib exhibited antitumor activity in cultured cells and xenograft tumor models harboring BRAF V600E and V600D mutations, and in a xenograft model harboring a BRAF fusion."	FDA / Day One	label — USPI §12.1, rev. 8/2025	"Paradoxical", "dimer", "squamous cell carcinoma" and "keratoacanthoma" are absent from the entire US label — verified absence, not a search miss. §12.2 contains no MAPK pharmacodynamic claim	supported
T2	"…including RAF monomers and dimers and BRAF fusion, suppressing activation of the MAPK pathway (see section 5.3)"	Ipsen / EMA	label — EU SmPC §5.1	CHMP grounds this on crystallography plus single-isoform enzymatic IC50s — not on any submitted head-to-head monomer-vs-dimer potency comparison. Note the cross-reference from §5.1 straight to §5.3, the NF1 ERK-activation text: the EU label internally qualifies its own claim. "Paradox", "squamous cell carcinoma" and "keratoacanthoma" return zero hits across the SmPC	partially-supported
T3	"In vitro, tovorafenib increased phosphorylation of ERK at clinically relevant concentrations in cells with NF1-loss of function suggesting activation, rather than inhibition, of the MAP kinase pathway. In an NF1 GEMM of plexiform neurofibroma without BRAF alteration, tovorafenib did not have antitumor activity, and while not statistically significant, an increase in tumor volume was noted in 2/12 mice (approximately 17%)."	FDA / Day One; EMA / Ipsen	label — USPI §13.2 + §5.6; EU SmPC §5.3 + §4.4	Day One in-house MSD pERK ELISA in sNF96.2, MeWo, NCI-H1838 at 1 h and 6 h; Nf1-flox/flox;Postn-Cre+ GEMM. Regulator-endorsed, label-tier documentation of MAPK activation by tovorafenib in BRAF wild-type, RAS-pathway-activated cells. Neither label uses the word "paradoxical"; both describe the phenomenon	supported
T4	"This effect is suggestive of paradoxical activation of the pathway in NF1-LOF tumour cell lines… paradoxical activation by tovorafenib at subtherapeutic doses (0.01–0.1 µM) was also described in literature. This could be linked to tovorafenib being ARAF sparing. Therefore, the limited non-clinical data do not support consistent activity of tovorafenib, and paradoxical activation could be an issue in this setting."	CHMP rapporteurs	regulatory-assessment — EPAR EMA/67438/2026 p.44 (NOT label)	A regulator, reasoning on the sponsor's own data, expressly contradicts the unqualified anti-paradox claim in the RAS-activated setting. None of this language reached either approved label	supported, at regulatory-assessment tier
T5	"Under identical conditions, paradoxical activation is not observed in cells treated with type II inhibitors at therapeutic concentrations" — but, in the same paper, "Both MLN2480 and AZD628 give some paradoxical activation at subtherapeutic doses (0.01–0.1 µM)"	Sun Y et al. (Dana-Farber; non-sponsor)	peer — Neuro Oncol 2017;19:774–785	Immunoblot pERK time-courses in murine neural progenitors transduced with HRAS-G12V; HEK293 activator/receiver constructs showing catalytic suppression, not prevention of dimerisation. This is the originating primary evidence for the entire anti-paradox narrative, and it is dose-conditional. 2024 corrigendum: vinculin loading control used twice across Fig 5A/5C, panel A replaced; undeclared borosilicate-glass handling ("high affinity for plastic that can confound the replication of EC50s"). Authors state central conclusions remain valid. Independent replication of the HRAS-G12V paradox experiments has not been published	partially-supported
T6	"MLN2480 binds with equal affinity to BRAF monomers and to preformed BRAF dimers"; "an equipotent antagonist of BRAF V600E, KIAA1549:BRAF, and other noncanonical BRAF oncoproteins that function as dimers"	Sun 2017	peer	Not a direct binding measurement to preformed dimers. Supported by matched cellular potency (190 nM fusion vs 248 nM V600E) and activator/receiver constructs. The Clayton corroboration is withdrawn — Clayton predicts equipotency from crystal geometry and defers to prior experiment. Tkacik's biochemistry is in mild tension with strict equipotency (dimer-favouring on Ki)	partially-supported
T7	"In contrast to type I BRAF inhibitors, tovorafenib did not induce paradoxical activation of the MAPK pathway in these models"	Kilburn LB et al. (Day One-funded, Day One co-authors)	peer — Nat Med 2024;30(1):207–217	A preclinical claim carried into the pivotal registrational clinical paper with no clinical pharmacodynamic support in FIREFLY-1 itself — no skin or tumour pERK collected. If the Sun citation is confirmed, it drops Sun's own subtherapeutic-dose qualifier	partially-supported
T8	Tovorafenib does not cause paradoxical activation in BRAF-fusion tumours — and, in the same paper, does raise pERK in NF1-LOF lines	Rastogi S et al. (all six authors Day One employees/stockholders)	peer — Cancer Res Commun 2025;5(4):668–679	Anti-paradox half scoped to BRAF fusions and supported there (AGK::BRAF PDX regression; pERK suppression). Paradox-positive half: MSD pERK ELISA — "Increased pERK levels were observed at lower concentrations of tovorafenib and decreased pERK levels at higher tovorafenib concentrations" — a bell-shaped curve. Vemurafenib comparator showed a monotonic rise in all three lines. Tovorafenib inactive in 2/3 NF1-LOF lines. Synergy with pimasertib: Loewe 39.8 / Bliss 19.0 (sNF96.2); Loewe 27.2 / Bliss 29.8 (ERMS PDX ex vivo)	partially-supported — the claim is true as scoped (BRAF fusions) and false as generalised (RAS-pathway-activated, BRAF wild-type)
T9	"In line with preclinical data suggesting that tovorafenib does not trigger paradoxical activation of ERK signaling, squamous cell carcinoma of the skin was reported as a TEAE in only 1 (<1%) of 149 treated patients"	Rasco DW et al. (Millennium/Takeda-sponsored)	peer — Cancer Chemother Pharmacol 2023;92:15–28	The 1/149 number is verified. Keratoacanthoma and new primary melanoma are NOT REPORTED — absence of report, not counted zeros. Uncontrolled absence-of-signal, open-label, no dermatological surveillance protocol, no comparator, short median exposure. The QW cohort's 12% (H-score) / 8% (image analysis) median tumour pERK reduction at day 21 means low target occupancy is an untested alternative explanation	partially-supported (downgraded)
T10	Positive binding cooperativity across RAF dimers (drug in one protomer increases affinity for the second)	Tkacik E et al. (Eck lab; NIH-funded; Eck holds a Novartis consultancy AND Novartis sponsored research support — Novartis sponsors the head-to-head comparator naporafenib)	peer — J Biol Chem 2023;299:104634	Full Hill-slope set: ARAF-SSDD −1.12; BRAF-WT −2.89; BRAF V600E −0.75; CRAF-WT −1.55 ± 0.43; CRAF-SSDD −3.18. Only two of the four non-V600E constructs show steep slopes; CRAF-WT sits nearer the monomeric controls. Purified-protein only; no cellular pERK data in RAS-mutant or BRAF wild-type cells anywhere in the paper. Also identifies the countervailing liability: IC50 >3000 nM against ARAF	partially-supported (downgraded)
T11	"In contrast to type I BRAF inhibitors, tovorafenib does not induce RAS-dependent paradoxical activation of the MAPK pathway"	Day One (Landi, Ziegler, Blackman, Cox et al.)	conference — ASCO 2022 TPS10062	Reference list extracted in full: the paradox bullet carries no superscript marker. Ref 6 (Sun 2017) is attached to the next bullet. The claim is broader than Sun supports and adds a "RAS-dependent" qualifier that no cited experiment tested for tovorafenib in a RAS-mutant system	ASSERTED-ONLY
T12	NF1 GEMM tumour increase in 2/12 mice	Bessler W et al. (incl. Day One employees)	conference — 2024 Global NF Conference poster	Never retrieved at source by anyone in this audit chain — reaches the record only via a CHMP footnote and Rastogi 2025. The three documents describe the endpoint inconsistently: tumour volume (labels) / tumour number (CHMP p.44) / "nine of 48 proximal nerve volume measurements" (Rastogi). A paradox-positive conference claim from the sponsor's own laboratory that reached both labels via the dossier	partially-supported (downgraded)
T13	"As of the clinical cutoff date, no patients had events meeting 'non-hematological malignant tumors' search criteria. No patients had any events of squamous cell carcinoma or keratoacanthomas."	CHMP reporting the sponsor's analysis	regulatory-assessment — EPAR p.115; FIREFLY-1 Arms 1+2, N=137, MedDRA v23.1, cutoff 10 May 2024	A prospectively pre-specified AESI with a defined MedDRA search — methodologically stronger than incidental TEAE tabulation. But secondhand, in a paediatric population with low baseline cuSCC risk, and with NF1 patients excluded — so it does not probe the RAS-activated setting where the preclinical signal sits	supported, at regulatory-assessment tier
T14	Three-year FIREFLY-1 follow-up	Day One	company / conference — SNO 2025, median study duration 40.6 months, cutoff 6 June 2025	Discloses median retreatment duration 9 months and median maximum tumour reduction −38.3%. The cuSCC / KA / SPM denominator at the 3-year cutoff is not disclosed; the "no new safety signals" characterisation was not found in the retrieved release and is unverified	UNKNOWN
T15	"About OJEMDA" boilerplate — label-derived, makes no paradox claim and no monomer/dimer claim	Day One	company	Notable for what it does not say: the sponsor's approved-product marketing language is more conservative than its scientific-congress language. The unqualified paradox claim lives in posters, protocols and reviews, not in the boilerplate	awareness only
T16	"highly-selective"	Day One	company	CHMP EPAR p.44 records inhibition of 14 human kinases at 0.2 µM, below unbound Cmax (Arg, Abl, DDR2, EphA1/A2/A8, Fyn, Lck, Lyn, MuSK, PTK5, SAPK2a, SAPK2b, plus transformed cKit-V560G and PDGFR-V561D), and that "The choice to determine IC50 for 10 out of 14 kinases is questionable." Also Adenosine A2A, A3 and NMDA/polyamine	partially-supported
T17	Tovorafenib "shows superior central nervous system penetration without the paradoxical activation of the MAPK pathway reported for type I BRAF inhibitors"	Expert Rev Clin Pharmacol 2024;17(11):999–1008 (and three further reviews)	review	No new experiments. Presents the claim unconditionally, stripping both Sun's concentration qualifier and the NF1-LOF exception both approved labels now carry. The clearest instance in the record of a claim hardening as it moves away from primary data	ASSERTED-ONLY
T18	No paradox index exists for tovorafenib	Adelmann 2016	peer (verified absence)	MLN2480, TAK-580, DAY101 and tovorafenib do not appear anywhere in the paper — verified, not inferred. Recorded so no reader imports a tovorafenib paradox index from this source	UNKNOWN

Biochemistry, two sources, unreconciled: Day One (Rastogi Table 2) enzymatic IC50 — BRAF V600E 7.1 nM, WT BRAF 10.1 nM, WT CRAF 0.7 nM, ARAF 55 nM. CHMP reproduced only three ("However, activity on ARAF was not presented"). Tkacik purified constructs — ARAF-SSDD IC50 >3000 nM / Ki >40 nM; BRAF-WT 633 nM / Ki 6.13; BRAF V600E 495 nM / Ki 20.5; CRAF-WT 94.2 nM / Ki 1.03; CRAF-SSDD 84.5 nM / Ki 0.21. On Ki this is dimer-favouring, not strictly equipotent. ARAF potency is unreconciled by more than 50-fold, and ARAF sparing is the mechanism both the sponsor and CHMP invoke to explain the NF1-LOF pERK rise.

STRUCTURAL COMPARISON

Molecule	Binding mode (and who says so)	DFG / αC / R506	Named contact residues	Co-crystal (PDB)	Monomer vs dimer selectivity
Vemurafenib	Type I½ conventionally; primary source's own term is αC-OUT (Karoulia 2016, peer). Neither label states any binding mode	DFG-IN / αC-OUT; R506 side chain "in a further OUT position"	L505, R506, R509 discussed as native and unmutated in the construct. ~~R506 salt bridge~~ STRUCK — review artefact	3OG7, 2.45 Å, deposited 2010-08-16. Two chains (A, B) in the ASU, resolving into two biological assemblies each annotated monomeric (C1/A1). Deposited entity is an "AKAP9-BRAF fusion protein" with ≥15 engineered solubilising mutations (I544A, I551K, Q562R, L588N, K630S, F667E, Y673S, A688R, L706S, Q709R, S713E, L716E, S720E). Entry does not annotate αC conformation. (PLX4720's 3C4C is frequently and incorrectly cited for vemurafenib)	Monomer-selective. >30-fold higher concentrations needed against p61 V600E dimers than R509H monomers (Yao 2015, cellular). No purified-dimer biochemistry exists. Purified catalytic-domain (monomer) enzymology exists: 35 / 110 / 48 nM (BRAF V600E / WT BRAF / CRAF)
Dabrafenib	αC-OUT (Karoulia 2016). "Type I½" is review-tier taxonomy applied inconsistently — record the structural observation, not the label. Neither label states any binding mode	αC-OUT overall, but R506 in a further IN position than PB and VEM, closer to the αC-IN inhibitor position	R506 only. No residue-by-residue contact map verified — unknown	4XV2 — "B-Raf Kinase V600E oncogenic mutant in complex with Dabrafenib", ligand P06, biological assembly a homodimer (C2, A₂) with dabrafenib bound in both chains. Also 5CSW (15 engineered mutations, E. coli), 6V2U (insect-cell WT BRAF, 3.78 Å), 5HIE (β3-αC loop deletion mutant, four chains)	Monomer-selective. 30–100× less potent against dimeric than monomeric BRAF V600E — but the same sentence reports this for PB and VEM too; it is not dabrafenib-discriminating. Counter-intuitively, dabrafenib promotes RAF priming and dimerisation more potently than VEM and PB. No sponsor dimer IC50 or monomer:dimer ratio has ever been published
Encorafenib	Type I½, DFG-in / αC-out (Marini 2022, peer). ATP-competitive per EU SmPC	DFG-in / αC-out. R506 not addressed	Docked pose, not experimental: H-bonds to Asp594 and Phe595 (DFG segment), Lys483 to the sulfonylamide; 1–3 H-bond with the Cys532 backbone in the hinge	NONE. Marini verbatim: "no experimental structure of encorafenib in BRAF is available." RCSB full-text search for "encorafenib" returns total_count = 1 — entry 9BP8, in which encorafenib is not the ligand. The pose is docked onto the BRAF–dabrafenib co-crystal 5CSW after self-docking validation	Monomer-selective by class assignment only. Pickles 2020's "group 1 / negative co-operativity" is re-cited from refs [4]–[6], not that paper's result. No side-by-side monomer vs dimer IC50, no dimer-selectivity ratio, no measured cooperativity coefficient. Functional dimer behaviour: 1 µM for 1 h inhibits pERK but activity recovers after washout "due to half-occupied active dimers" (Cotto-Rios 2020)
Mosperafenib	NOT DISCLOSED. No type I / I½ / II designation anywhere. A specialist review tabulating this field for every comparator paradox breaker entered "NA" for this compound	Not disclosed. No DFG, αC or R506 status	None disclosed	NONE. No co-crystal, no PDB ID	Not publicly readable. No forced-dimer comparison, no selectivity ratio, no co-IP/BRET/NanoBiT/crosslinking. A review's dimer column headed "confirmed experimentally" enters BRAF/BRAF, citing two paywalled 2022 papers — so data may exist behind the paywall. Independent review table and vendor Kd both describe a compound engaging wild-type BRAF at least as tightly as V600E (review: BRAF WT <5E-10 M vs V600E 1.77E-09 M; vendor Kd 0.6 / 1.2 / ~1.7 nM for WT BRAF / V600E / CRAF) — a pan-RAF-like binding profile nobody has reconciled with the paradox-breaker positioning
NXP200	NOT DISCLOSED. No type designation, no DFG or αC configuration, no SPR/ITC, no biophysics of any kind	Not disclosed	None disclosed	NONE. No PDB ID. The chemical structure itself is not public — PubChem PUG-REST returns "No CID found" for both HSK42360 and NXP200; ChEMBL returns 0 records; no attributable Haisco patent	Not disclosed. No monomer-vs-dimer potency pair, no CRAF or heterodimer IC50, no biochemical dimer-affinity measurement. Sole potency value in the entire record: BRAF V600E enzymatic IC50 5 nM. Whether the mechanism is monomer selectivity or dimer binding cannot be determined from the public record
Claturafenib	"Orthosteric" (ATP-site) per the peer source. No type I / I½ / II designation is made by any source. αC-OUT explicitly reported; DFG-in vs DFG-out is never stated — no DFG terminology appears	Ordered A-loop; αC rotated outward. DFG state unassigned	NOT DISCLOSED in any accessible source	9BP8, 1.73 Å, deposited 2024-05-07, released 2025-05-14, ligand A1AQY, single chain (~294 residues, matching the BRAF-KDL 432–726 construct), depositor T.-C. Mou = co-author on the primary paper. Primary citation "To be published" — an unrefereed database deposit, NOT peer tier. The paper's Data Availability carries no accession. The deposited asymmetric unit is a monomer and therefore demonstrates nothing about dimers	No head-to-head biochemical monomer-vs-dimer comparison exists. No dimer-selectivity ratio is reported anywhere. Target engagement (ITDR/TMT in A375 lysates): "bound BRAF^V600E with a 10-fold higher affinity compared with wild-type CRAF protein" — a mutant-vs-wild-type-protomer ratio, not a monomer-vs-dimer ratio. Cell pERK IC50: class I 0.7–7 nM; class II 10–14 nM; class III 0.8–7.8 nM; indels 113–179 nM; BRAF wild-type ≥9,800 nM (a lower bound: ≥1,400× to ≥14,000×)
Tovorafenib	Type II — grounded at label tier in BOTH jurisdictions (USPI §12.1; EU SmPC §5.1). The only molecule in this set whose binding mode is label-stated	DFG-OUT / αC-IN (Tkacik 2023). CHMP concurs: displaces the phenylalanine out of the back pocket, shifts the C-helix, changes the glycine-rich loop conformation; "similar to sorafenib and differs from the Type-I BRAF inhibitor vemurafenib"	C532, T529, K483, D594, E501 — confirmed at peer tier	6V34 (BRAF V600E + tovorafenib), 8F7O (WT BRAF + tovorafenib), 8F7P (WT BRAF + naporafenib). Both protomers of a back-to-back kinase-domain dimer occupied with identical binding mode — the only molecule in this set with a published drug-occupied dimer structure	Claimed equipotent monomer + dimer, at label tier (EU SmPC §5.1 only). No head-to-head cellular monomer-vs-dimer potency ratio disclosed by anyone. Supporting: matched cellular potency (KIAA1549:BRAF 190 nM vs V600E 248 nM); activator/receiver constructs; both-protomer occupancy in crystal. Against strict equipotency: Tkacik Ki is dimer-favouring (~3.3× on the WT BRAF dimer, ~20–100× on CRAF constructs, vs monomeric BRAF V600E). Positive cooperativity claimed from Hill slopes — but only 2 of 4 non-V600E constructs are steep, and CRAF-WT (−1.55) sits nearer the monomeric controls

Cross-cutting structural conclusion. Binding class does not predict paradox in either direction. Tkacik 2026 (eLife RP110344) reports that type I and type II inhibitors induce paradoxical activation of all three RAF isoforms, with the type II inhibition phase suggestive of positive cooperativity — in conflict with the prevailing negative-allostery model. Any argument of the form "molecule X is class Y, therefore paradox / no paradox" should be retired, not caveated.

04By molecule

Each audit in full, after adversarial re-verification. Click to expand.

claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026

Verification confidence: high

Claturafenib — Paradox-Claim Audit, Adversarially Corrected

Data cut: 1 September 2026. Scope: public-domain evidence only. Plixorafenib (Fore's own asset) is out of scope and is not researched, characterised or compared anywhere below; where it appears as a comparator in a source, it is redacted as [out-of-scope comparator — handled by the separate cleared process].

Verification status of this document: the underlying audit was re-checked against primary sources on 1 September 2026. Full text of Yaeger et al. was read at PMC11372368; NCT05355701 was pulled from the ClinicalTrials.gov API v2; PubMed, PubChem and RCSB were retrieved directly. Two sources could not be opened (see §7). Fourteen corrections and seven venue/verdict downgrades were applied.

1. Identity — RESOLVED (re-verified)

Field	Value	Source
INN	claturafenib	Multiple; not indexed in PubMed under this INN
Codes	PF-07799933 (Pfizer); ARRY-440 (Array legacy)	Yaeger 2024, verbatim
CAS	2754408-94-9	PubChem, vendor monographs
PubChem CID	165150001	PubChem REST, retrieved 1 Sep 2026
Formula / MW	C18H15Cl2F2N5O3S / 490.3	PubChem
IUPAC	N-[2-chloro-3-[(5-chloro-3-methyl-4-oxoquinazolin-6-yl)amino]-4-fluorophenyl]-3-fluoroazetidine-1-sulfonamide	PubChem
PDB ligand	A1AQY	RCSB 9BP8
Originator	Array BioPharma — verbatim: "PF-07799933 (ARRY-440), an orthosteric, pan-mutant BRAF inhibitor, was discovered by Array BioPharma."	Yaeger 2024 (peer)
Sponsor	Pfizer Inc. (Array acquired 2019)	company
Stage	Investigational, Phase 1/1b — NCT05355701 (n=103 actual, Active-not-recruiting, est. primary completion 31 Aug 2027, hasResults=FALSE) and NCT05538130 (with polfurmetinib / PF-07799544, brain-penetrant MEK inhibitor)	registry

CORRECTED — removed: IUPHAR/BPS ligand ID 13808. The GtoPdb page returns a login screen and the identifier could not be confirmed. The original audit asserted it as resolved while its own citation block admitted the page was never read. [FLAG: re-establish or drop the IUPHAR ID.]

Companion molecule confirmed: PF-07799544 = polfurmetinib, oral brain-penetrant reversible MEK1/2 inhibitor (Pfizer Oncology Development; ASCO 2026 abstract title).

Registry-class note. No public source in any tier applies the term "paradox breaker" to claturafenib. Re-confirmed: the Cancer Discovery paper never uses the phrase; Pfizer's own descriptors are "next-generation BRAF dimer inhibitor" (ASCO 2026 title) and "pan-mutant BRAF inhibitor". The Fore CI registry class is an internal attribution with no public venue and must not be tagged "company".

2. Label tier — EMPTY (verified by retrieval, not by silence)

No FDA USPI and no EMA SmPC exist. Verified three ways: (a) PubMed returns exactly two records for PF-07799933 / ARRY-440 / claturafenib — the Yaeger primary paper and the Scotece & Drosten commentary; (b) no Drugs@FDA or EMA product entry; (c) every source, including Pfizer's own page and the aggregator blog, states the molecule is investigational and unapproved.

Zero label-grade evidence supports any paradox claim about this molecule, and none can exist until approval. This is a finding, not a gap in searching.

3. Binding mode — PARTIALLY DISCLOSED

Class of binding. "Orthosteric" (ATP-site) per the peer source; vendor copy adds "ATP-competitive" (no tier). No type I / I½ / II designation is made by any source.
Conformation. Verbatim, corrected to full: PF-07799933 "induced an ordered A-loop and caused an outward rotation in the αC-helix, both consistent with an inactive kinase conformation that may destabilize RAF dimers or prevent dimerization." — αC-OUT is explicitly reported. DFG-in vs DFG-out is never stated; no DFG terminology appears.

Analyst inference, not a source claim: an orthosteric, αC-OUT, A-loop-ordered pose is the signature usually classed type I½ / αC-OUT (type IIA). No source makes this assignment. Recorded as UNASSIGNED. [FLAG: do not let this inference travel without this disclaimer.]
Crystallography. Two constructs: BRAF-KDL (residues 432–726) and BRAF-KDS (residues 445–723, three solubilising point mutations). Molecular replacement from PDB 4MNE via PHASER. (Corrected — the original audit named only KDL and attributed the work to PHENIX/COOT.) [FLAG: confirm refinement software against Supplementary Methods.]
PDB 9BP8 — "Crystal structure of BRAF kinase domain with PF-07799933", X-ray 1.73 Å, deposited 7 May 2024, released 14 May 2025, single chain (~294 modelled residues), ligand A1AQY, depositor T.-C. Mou, primary citation "To be published". CORRECTED (linkage upgraded): depositor T.-C. Mou is co-author Mou TC on Yaeger 2024, affiliated to Pfizer Boulder Research and Development; and the single-chain entity length matches the BRAF-KDL 432–726 construct (295 residues). The 9BP8 ↔ paper linkage is strongly corroborated by shared authorship and construct length, though still not formally established: the paper's Data Availability statement carries no accession code (verbatim: "Upon request, and subject to review, Pfizer, Inc. will provide the data that support the findings of this study…"), and 9BP8's own citation remains unpublished. 9BP8 is an unrefereed database deposit — it is not "peer" tier. The deposited asymmetric unit is a monomer and therefore demonstrates nothing about dimers.
Key contact residues — NOT DISCLOSED in any accessible source.
STRUCK — overreach. The original audit's note that the ligand's 3-fluoroazetidine-1-sulfonamide is "of the general class described in kinase-modulator patent literature as dissociating pathway inhibition from pathway activation" is removed. It is a structure-to-behaviour inference with no source tied to this molecule and functions as a back-door paradox argument built from chemistry. The chemical fact (the sulfonamide substituent, confirmed in the PubChem IUPAC name) stands alone with no behavioural gloss.

4. Monomer vs dimer — NO HEAD-TO-HEAD BIOCHEMICAL COMPARISON EXISTS

No biochemical assay compares IC50 against monomeric RAF versus enforced/constitutive RAF dimers. No dimer-selectivity ratio is reported anywhere. All disclosed data are from Yaeger et al., Cancer Discov 2024;14(9):1599–1611 (PMID 38691346; DOI 10.1158/2159-8290.CD-24-0024):

(a) Cell-based pERK IC50 panel — quantitative in-cell Western, 19 cancer cell lines (corrected: not all patient-derived; MEL21514 is patient-biopsy-derived, others are established repository lines such as A375 and HT29), 1-hour dose-response, four-parameter fit:

Class	IC50
Class I (V600E/K, monomer-driven)	0.7–7 nmol/L
Class II (L597V/R, G469A, K601E — constitutively dimeric)	10–14 nmol/L
Class III (G466V, D594G — kinase-impaired, transactivating)	0.8–7.8 nmol/L
BRAF indels (V487-P492, N486-P490, L485-P490)	113–179 nmol/L
Acquired BRAFV600E/p61 splice variant (MEL21514)	59 nmol/L
Acquired NRAS-Q61K on BRAFV600E (A375-NRASQ61K)	16 nmol/L
BRAF wild-type	≥9,800 nmol/L

Comparators run head-to-head: encorafenib, belvarafenib, tovorafenib, naporafenib, exarafenib, and [out-of-scope comparator — handled by the separate cleared process].

CORRECTED — deleted: the original audit's "roughly a 2–20× potency loss on dimeric Class II relative to monomeric Class I." That is arithmetic across non-isogenic cell lines, computed in the same paragraph that states no such comparison exists. It is not a proxy measurement and must not be quoted as one. Wild-type sparing is stated correctly as a lower bound only: ≥9,800 nmol/L implies ≥1,400× to ≥14,000× relative to Class I lines, not a range.

(b) Target engagement. ITDR with TMT mass-spec in A375 BRAFV600E lysates, 0.122–2,000 nmol/L: PF-07799933 "bound BRAFV600E with a 10-fold higher affinity compared with wild-type CRAF protein." This is a mutant-vs-wild-type-protomer affinity ratio, not a monomer-vs-dimer ratio.

(c) Dimer-disruption co-IP — NON-MATCHED SYSTEMS. Mutant arm: endogenous RAF co-IP in MEL21514 and A375-NRASQ61K, 1 h, Western — BRAFV600E:wild-type-CRAF heterodimers "were disrupted". Wild-type arm: HEK293T over-expressing exogenous GFP/V5-tagged BRAF and CRAF — PF-07799933 "failed to disrupt wild-type-BRAF:wild-type-CRAF dimer complexes." The two arms are not matched systems; no endogenous wild-type dimer experiment is reported.

"Equipotent against monomer and dimer" is claimed by no one and is not supported.

5. Behaviour by alteration

Class I (V600E/K) — strongest data. pERK IC50 0.7–7 nmol/L; xenograft regressions incl. deeper regressions than encorafenib+binimetinib in A375; clinical confirmed responses systemically and intracranially in patients refractory to approved RAF inhibitors. ASCO 2026 abstract 9512 (with polfurmetinib): ORR 27% (41 dose-escalation) and 32% (19 dose-optimisation) in advanced melanoma; intracranial ORR 30% (23 pts) and 22% (9 pts).
Class II (G469A, K601E, L597V/R) — preclinical pERK IC50 10–14 nmol/L; regressions in BRAF G469A NSCLC and BRAF K601E melanoma xenografts. Clinically unproven (below).
Class III (G466V, D594G) — preclinical pERK IC50 0.8–7.8 nmol/L. Clinically unproven.
Class II/III clinical reality: among eight patients (4 Class II, 4 Class III) on monotherapy or with binimetinib/cetuximab, "there were no responses" by RECIST 1.1 / RANO. Only one received the efficacious 225 mg BID + binimetinib dose; that patient (BRAF G466E breast adenoid cystic carcinoma) had a ctDNA molecular CR with radiographic progression. The "pan-mutant" framing is preclinically grounded, clinically unconfirmed for the dimeric and transactivating classes.
BRAF indels — pERK IC50 113–179 nmol/L (an order of magnitude weaker than Class I/III), but tumour regressions were reported in a BRAF-indel pancreatic cancer xenograft (added — omitted from the original audit).
BRAF fusions (KIAA1549-BRAF etc.) — NO PRECLINICAL DATA. No fusion-positive patient among the 30 reported. CORRECTED to unknown, not absent: NCT05355701 closed at n=103 with hasResults=FALSE, and its eligibility text ("BRAF alterations… V600 and Class II/III") neither explicitly includes nor excludes fusions, so fusion enrolment among the 73 unreported patients is unknown.
Acquired resistance — BRAFV600E/p61 splice variant 59 nmol/L; A375-NRASQ61K 16 nmol/L; one BRAF-inhibitor-refractory papillary thyroid cancer with a BRAF p48 splice variant on 150 mg QD monotherapy, progressed, then achieved sustained −80% reduction after binimetinib was added.
RAS-mutant / BRAF-wild-type context — the context that defines paradoxical activation: ESSENTIALLY UNTESTED. The only RAS-mutant model, A375-NRASQ61K, is a BRAFV600E melanoma line engineered with acquired NRAS-Q61K — it is not a BRAF-wild-type/RAS-driven paradox model. No KRAS/NRAS-mutant BRAF-wild-type line, xenograft or patient cohort. No keratinocyte, skin or 3D skin-equivalent model. No in vivo paradox experiment of any kind. CORRECTED: the original audit stated "the trial excluded RAS-mutant patients." NCT05355701 contains no RAS/KRAS/NRAS exclusion criterion. Such patients are absent because enrolment requires a qualifying BRAF alteration — an inclusion-criterion consequence, not a stated exclusion. [FLAG: correct anywhere this was propagated downstream.]
BRAF wild-type — pERK IC50 ≥9,800 nmol/L (no inhibition) and, critically, "less paradoxical activation of pERK than encorafenib" — reduced but not absent. Magnitude undisclosed (see §6, claim 1).

6. Paradox claims — venues re-tagged, verdicts re-checked

C1 · "In BRAF wild-type cells, PF-07799933 demonstrated no pERK inhibition, in contrast to the pan-RAF dimer inhibitors, and less paradoxical activation of pERK than encorafenib."

Who: Yaeger R et al.; senior author Rothenberg SM (Pfizer). Venue: peer ✔ (verified in original, PMC11372368).
Source: Cancer Discov 2024;14(9):1599–1611. DOI 10.1158/2159-8290.CD-24-0024; PMID 38691346; PMC11372368.
Evidence: quantitative pERK in-cell Western dose-response, 19-line panel including BRAF wild-type lines, 1 h, four-parameter fit, head-to-head comparators as listed in §4. Wild-type IC50 ≥9,800 nmol/L.
DOWNGRADED evidence note: the claim is explicitly comparative and relative — less paradoxical activation, not absence. No fold-induction value, induction EC50 or paradox index is reported in the text, so "less than encorafenib" cannot be sized. (Corrected from "NOT quantified": the wild-type dose-response resides in Supplementary Fig. S1A, which was not inspected. An absence claim about an un-opened figure is itself a silence-for-absence error.) [FLAG: open Supplementary Fig. S1A before any external use.] The assay is a 1-hour in vitro pERK readout in BRAF wild-type lines — not a RAS-mutant paradox model, not in vivo, not a skin model.
Verdict: partially-supported.

C2 · "…disrupted endogenous mutant-BRAF:wild-type-CRAF dimers, and spared wild-type ERK signaling."

Who: Yaeger, Rothenberg et al. Venue: peer ✔.
Evidence: two co-IP experiments in non-matched systems (endogenous mutant arm vs HEK293T over-expression wild-type arm), plus ITDR 10-fold BRAFV600E-over-CRAF affinity. No endogenous wild-type dimer experiment; no functional RAS-driven wild-type dimer readout.
Verdict: partially-supported.

C3 · Cocrystal conformation explains paradox avoidance

Who: Yaeger, Rothenberg et al. Venue: peer — for the Cancer Discovery sentence only. PDB 9BP8 is an unrefereed deposit ("To be published") and is separated out of the peer source line.
Evidence: one X-ray cocrystal of BRAF-KDL with PF-07799933 (MR from 4MNE via PHASER). Structural statistics relegated to Supplementary Table S2 (not accessed); Data Availability carries no accession. The mechanistic link is hedged by the authors themselves — "may destabilize RAF dimers or prevent dimerization." No biophysical dimer-affinity measurement (SPR, SEC-MALS, AUC, cross-linking) accompanies it. 9BP8 is a monomer.
Verdict: partially-supported.

C4 · "Since it does not affect wild-type (WT) dimers, it has a favorable safety profile in the clinic."

Who: Scotece M, Drosten M (CIC/IBMCC, CSIC-Universidad de Salamanca) — independent commentators, not the sponsor. Venue: review ✔ (correctly tagged in the original).
Source: Trends in Cancer 2024 Jul;10(7):576–578. DOI 10.1016/j.trecan.2024.05.009; PMID 38866669.
Evidence: none of its own. The causal bridge (wild-type dimer sparing → favourable clinical safety) is not established by the underlying paper. The clinical dataset is 30 Phase 1 patients in whom the paradox-toxicity endpoint was never affirmatively measured or reported. The commentary also restates the wild-type dimer finding as fact without the over-expression caveat present in the source.
Verdict: asserted-only. A review can never on its own establish that a molecule does or does not cause paradoxical activation.

C5 · "Works by disrupting the formation of BRAF-containing dimers" / "highly selective for mutant BRAF, significantly sparing normal (wild-type) cells to reduce off-target side effects."

Who: New Drug Approvals, self-published drug-monograph blog, 19 Apr 2026.
VENUE DOWNGRADED: "review" → AGGREGATOR/BLOG (lowest awareness tier, no editorial or peer process). A self-published monograph must not share a tier with a Trends in Cancer commentary. Verified on retrieval: the page cites no primary literature — only NCT numbers and patent numbers.
Evidence: none. Materially inverts the mechanism: the primary data show disruption of mutant-BRAF-containing dimers and failure to disrupt wild-type BRAF:CRAF dimers, the latter being the proposed basis of wild-type sparing. An unqualified "disrupting the formation of BRAF-containing dimers" is not what the evidence shows.
Verdict: contradicted.

C6 · "Next-generation BRAF dimer inhibitor" (sponsor's own descriptor)

Who: Pfizer Inc.; presenting author Monica Chen, Memorial Sloan Kettering (presenter and session independently confirmed — Rapid Oral, Melanoma, Sat 30 May 2026).
Venue: conference ✔ — but split from the company press release, which is a separate company-tier item and should not sit inside a conference source line.
Source: ASCO 2026 Abstract 9512; J Clin Oncol 2026;44(16_suppl):9512.
Evidence: a descriptor in a title, not a data-backed claim. The sponsor does not use "paradox breaker" and makes no paradox claim in its own conference or press material. Reported content is clinical only: any-grade TEAEs 95% (rash 40%, fatigue 37%, diarrhoea 32%, peripheral oedema 22%); grade ≥3 67% (rash 7%, disease progression 7%, anaemia 5%, fall 5%); ORR 27%/32%; intracranial ORR 30%/22%. No paradox, dimer-selectivity or wild-type-sparing data are presented.
[FLAG: ascopubs.org returns HTTP 403 — re-confirmed 1 Sep 2026. All nine percentages were independently reproduced from secondary indexing and are almost certainly correct, but the abstract remains UNREAD IN ORIGINAL. Do not promote to verified without obtaining the PDF.]
Verdict: unknown (as a paradox claim).

C7 · "Selective for inhibition of mutated BRAF, which may contribute to a favorable therapeutic safety profile."

VENUE SPLIT (was one "company" tag over three different kinds of source):
company — pfizeroncologydevelopment.com/molecule/brafi. [FLAG: HTTP 403 on direct fetch, re-confirmed. Read only via search-index snippets — INCOMPLETELY VERIFIED.]
no tier — vendor catalogues (MedChemExpress, Selleck, TargetMol, MedKoo, InvivoChem, Probechem, AdooQ, Chemietek, PatSnap Synapse). Third-party commercial copy, not the sponsor, no evidentiary weight.
tertiary database — NCI Thesaurus-derived phrasing is a government reference work, not a corporate assertion.
CORRECTED — quote misattribution: "significantly sparing normal (wild-type) cells to reduce off-target side effects" is the New Drug Approvals blog's wording (C5), not Pfizer's, and was wrongly duplicated into this claim. [FLAG: search-index snippets blend adjacent sources; per-sentence attribution between the 403 Pfizer page and the blog is unreliable.]
Evidence: derivative restatement of the ≥9,800 nmol/L wild-type figure. No independent experiment. The safety inference is forward-looking, hedged, and tied to no reported cutaneous-toxicity dataset.
Verdict: asserted-only.

C8 · Classification as a "paradox breaker"

Who: Fore CI registry (internal). No external attributor identified.
VENUE DOWNGRADED: "company" → NO VENUE / INTERNAL ATTRIBUTION ONLY. The source field itself says "no public source"; tagging it "company" places an internal registry entry inside a defined public evidence tier, where downstream readers will mistake it for a corporate claim.
Evidence: none. Re-confirmed: the primary paper uses "paradox"/"paradoxical" in exactly two sentences — one generic background statement about approved RAF inhibitors, one relative claim ("less paradoxical activation… than encorafenib") that affirms residual paradoxical activation rather than its absence.
Verdict: asserted-only. [FLAG: if this class label is used downstream, it must carry "internal Fore attribution, unmirrored in the public record."]

C9 · Absence-of-paradox-toxicity by silence

VENUE DOWNGRADED: "peer" → REVIEW / INFERENCE. No peer source makes this claim; the peer source is silent. Only the Trends in Cancer commentary converts that silence into a safety conclusion. The venue of a claim is the venue of whoever makes it, not the venue of the document that fails to mention it — this is the single most consequential tagging error in the original audit.
Evidence: ABSENCE OF REPORT, NOT EVIDENCE OF ABSENCE. In Yaeger 2024 the AE reporting threshold was verbatim "TEAEs reported in ≥3 patients regardless of attribution are shown in Supplementary Table S4", and the safety section is simply silent on cuSCC / keratoacanthoma / new primary melanoma / skin papilloma / hyperkeratosis. No skin-surveillance protocol, dermatologic screening schedule, or explicit negative statement is reported. In ASCO 2026 the denominator is ~60 and rash is the leading TEAE (40% any grade, 7% grade ≥3) — but every patient received a concurrent MEK inhibitor, which independently suppresses paradox-driven cutaneous neoplasia [FLAG: this mechanistic premise is correct field knowledge but is uncited in this audit — attach a peer citation or mark it as background], invalidating that cohort as a paradox readout. NCT05355701 has posted no results.
Verdict: unknown.

7. Clinical paradox signal — NONE INTERPRETABLE, IN EITHER DIRECTION

(1) Yaeger 2024, Phase 1 — 30 patients treated, Aug 2022–Jul 2023, nine US/Canadian centres.

CORRECTED — the key denominator was wrong. Dose-level enrolment reads: 50 mg QD (n=4 mono); 150 mg QD (n=4 mono, n=4 +binimetinib, n=4 +cetuximab, all CRC); 225 mg BID (n=5 mono, n=4 +binimetinib); 450 mg BID (n=5 mono).

Monotherapy = 18. Combination = 12. Total = 30. The original audit reported the monotherapy safety denominator as 30. It is 18. [FLAG: derived by arithmetic from the dose-level string; it sums exactly to 30, but confirm against Table 1 before external use.]

No DLTs; MTD not reached. Monotherapy (n=18): any-grade TEAE 94%, grade ≥3 28% — fatigue 44%, headache 28%, blurred vision 22%, lipase increased 16%. Combination (n=12): any-grade 100%, grade ≥3 44% — peripheral oedema 33%, acneiform rash 28%, diarrhoea 28%, fatigue 28%. (Corrected: the original audit listed blurred vision among the combination TEAEs; it is a monotherapy event.) cuSCC, keratoacanthoma, new primary melanoma, skin papilloma and hyperkeratosis are not mentioned — neither reported nor explicitly excluded.

(2) ASCO 2026 Abstract 9512 — 41 dose-escalation + 19 dose-optimisation (~60), all on a concurrent MEK inhibitor. Cannot serve as a paradox readout at all.

(3) NCT05355701 — n=103 actual, Active-not-recruiting, hasResults=FALSE (API v2, 1 Sep 2026). No structured AE table is public, so no denominator-level cutaneous-neoplasia rate is verifiable at any dose or duration.

(4) Exposure duration — the variable governing whether paradox-driven cuSCC appears at all — not reported in any accessible source.

BOTTOM LINE (strengthened by correction): the entire public monotherapy safety denominator for this molecule is 18 patients, with a silent skin section, no dermatologic surveillance protocol, and unreported exposure duration. That is far too small and too briefly exposed to support or refute paradox avoidance. Any assertion that claturafenib's clinical safety demonstrates paradox breaking is currently UNFALSIFIABLE, not validated.

8. Unknowns (an honest unknown is a finding)

Label tier empty. Zero label-grade evidence exists and none can until approval.
No public source calls claturafenib a "paradox breaker." Sponsor's own term is "BRAF dimer inhibitor."
The primary paper makes a RELATIVE, not absolute, paradox claim — it concedes residual paradoxical activation. No source anywhere states "does not induce paradoxical activation."
Magnitude of residual paradoxical induction not reported — no fold-change, induction EC50 or paradox index in text. Supplementary Fig. S1A not inspected.
No paradox index. The molecule has never been run through a published paradox-index assay and cannot be placed on the standard comparative scale.
No RAS-mutant / BRAF-wild-type paradox experiment exists — no such cell line, no xenograft, no in vivo model, no skin/keratinocyte or 3D skin-equivalent, no patient cohort (absent by inclusion criterion, not by an explicit RAS exclusion). The single largest evidentiary gap: the defining context for paradoxical activation has not been tested.
No BRAF fusion data preclinically; fusion enrolment among the 73 unreported patients in NCT05355701 is unknown, not confirmed absent.
No monomer-vs-dimer biochemical potency comparison — no enforced-dimer assay, no selectivity ratio, no SPR/SEC-MALS/AUC/cross-linking. The mechanism rests on a hedged structural inference plus non-matched co-IP.
The wild-type dimer co-IP is an over-expression experiment; no endogenous wild-type dimer experiment is reported.
Binding-mode type not assigned; DFG state never stated; contact residues not disclosed.
PDB 9BP8 linkage corroborated but not formal — shared depositor/author and matching construct length; primary citation still "To be published"; no accession in Data Availability. No dedicated medicinal-chemistry or structural discovery paper exists.
No cutaneous-neoplasia rate at any denominator. Monotherapy denominator 18; AE threshold ≥3 patients; no surveillance protocol; NCT05355701 unposted.
The only larger safety dataset is confounded by a concurrent MEK inhibitor.
Exposure duration not reported — latency to paradox-driven cutaneous neoplasia cannot be assessed.
Class II/III clinical activity unproven — zero responses in eight patients; only one at the efficacious dose.
Sources not read in original (HTTP 403 / login / paywall), re-confirmed 1 Sep 2026: the ASCO abstract page (ascopubs.org), the Pfizer Oncology Development molecule page, the AACR article page, the Trends in Cancer full text, and the IUPHAR/BPS ligand page. Everything quoted from these is snippet-derived; search-index summarisers blend adjacent sources, which is how the C5/C7 quote duplication occurred. Re-verify verbatim before any external use.
NEW GAP — patent literature never searched. The aggregator blog references patent numbers; Array/Pfizer composition-of-matter and method-of-use filings routinely contain BRAF-wild-type pERK and paradox data that never reach a journal. This is an entire unexamined evidence stream. [FLAG: search the PF-07799933 / ARRY-440 patent family before the next revision.]
No independent (non-sponsor) experimental replication exists. The paper is sponsor-led with academic clinical co-authors (MSK and multi-site investigators), but every preclinical, structural, target-engagement and co-IP experiment is by Pfizer Boulder R&D authors. No academic group has independently profiled claturafenib for paradoxical activation. The compound is sold by at least eight commercial vendors, so independent profiling is feasible — its absence is a genuine gap, not an access constraint.

9. Citations

PEER (primary) — Yaeger R, McKean MA, Haq R, … Mou T-C, Wu W-I, … Rosen N, … Rothenberg SM. A Next-Generation BRAF Inhibitor Overcomes Resistance to BRAF Inhibition in Patients with BRAF-Mutant Cancers Using Pharmacokinetics-Informed Dose Escalation. Cancer Discovery 2024;14(9):1599–1611. DOI 10.1158/2159-8290.CD-24-0024; PMID 38691346; PMC11372368. Full text read in original. The ONLY primary peer-reviewed source for this molecule.
REVIEW (secondary, awareness only) — Scotece M, Drosten M. A new BRAF inhibitor breaks resistance barriers. Trends in Cancer 2024 Jul;10(7):576–578. DOI 10.1016/j.trecan.2024.05.009; PMID 38866669. [FLAG: full text not read — abstract/snippet only.]
CONFERENCE — Chen M, et al. ASCO 2026 Abstract 9512; J Clin Oncol 2026;44(16_suppl):9512. Rapid Oral, Melanoma, 30 May 2026. [FLAG: HTTP 403 — not read in original.]
COMPANY — Pfizer Inc. press release, "Pfizer Showcases Oncology Innovation and Next-Generation Pipeline at ASCO 2026," 20 Apr 2026.
COMPANY — Pfizer Oncology Development molecule page, claturafenib / PF-07799933. [FLAG: HTTP 403 — snippet-only, INCOMPLETELY VERIFIED.]
REGISTRY — ClinicalTrials.gov NCT05355701, Pfizer, Phase 1, Active-not-recruiting, enrolment 103 (actual), start 5 Jul 2022, est. primary completion 31 Aug 2027, hasResults = FALSE. Retrieved via API v2, 1 Sep 2026. No RAS/KRAS/NRAS exclusion criterion.
REGISTRY — ClinicalTrials.gov NCT05538130 (polfurmetinib / PF-07799544 platform study, includes the claturafenib combination arm).
STRUCTURAL (unrefereed deposit — NOT peer tier) — RCSB PDB 9BP8, X-ray 1.73 Å, deposited 7 May 2024, released 14 May 2025, ligand A1AQY, single chain, depositor T.-C. Mou, primary citation "To be published."
IDENTIFIER — PubChem CID 165150001 (retrieved in original); CAS 2754408-94-9. (IUPHAR/BPS ligand 13808 removed — unverifiable, page behind login.)
AGGREGATOR / BLOG (lowest awareness tier) — "Claturafenib," New Drug Approvals, 19 Apr 2026. Read in original; cites no primary literature; source of the mechanistically inverted "disrupting the formation of BRAF-containing dimers" statement.
SECONDARY REPORTING — Melanoma Research Alliance, "Highlights in Melanoma Treatment Advances from ASCO 2026."
NEGATIVE SEARCH RESULT (recorded as evidence) — no FDA USPI and no EMA SmPC exist as of 1 Sep 2026; PubMed returns exactly two records for this molecule. Verified by retrieval, not assumed from a missed search.

10. What did NOT survive re-verification

#	Original audit statement	Status
1	"Total public monotherapy safety denominator is 30 patients"	WRONG — it is 18
2	"the trial excluded RAS-mutant patients"	NOT IN THE RECORD
3	Verbatim cocrystal quote ending "…may destabilize RAF dimers"	TRUNCATED — "or prevent dimerization" dropped
4	Claim 9 tagged peer	DOWNGRADED to review/inference
5	Claim 8 tagged company	DOWNGRADED to internal, no venue
6	Claim 7 tagged company (one tag over three source kinds)	SPLIT: company / no-tier vendor / tertiary DB
7	Claim 5 tagged review	DOWNGRADED to aggregator blog
8	"significantly sparing normal (wild-type) cells…" attributed to Pfizer	MISATTRIBUTED — it is the blog's wording
9	IUPHAR ligand 13808 stated as resolved	UNVERIFIED — removed
10	"2–20× potency loss on dimeric Class II"	DELETED — arithmetic on incomparable lines
11	Sulfonamide → paradox-dissociation patent-class note	STRUCK — structure-to-behaviour overreach
12	"no fusion-positive patient enrolled in any reported cohort"	NARROWED — true for n=30; unknown for n=103
13	"19 patient-derived cell lines"	NARROWED — 19 lines, not all patient-derived
14	"magnitude NOT disclosed" (wild-type paradox)	SOFTENED — Suppl. Fig. S1A not inspected
15	"single Pfizer-authored paper"	REFINED — sponsor-led; all mechanistic work sponsor-generated

Everything else in the original audit survived re-verification. The identity resolution, the empty-label finding, the relative-not-absolute reading of the paradox sentence, the non-matched co-IP systems critique, the MEK-confounding of the ASCO cohort, and the "paradox breaker" class being an unmirrored internal attribution are all confirmed against primary sources.

Dabrafenib (TAFINLAR). Data cut: 1 September 2026.

Verification confidence: high

« # Dabrafenib (TAFINLAR) — Paradox-Claim Audit, ADVERSARIALLY CORRECTED

Data cut: 1 September 2026. Plixorafenib / FORE8394 / PLX8394 is out of scope for this file. No comparison is drawn, no property inferred, no ranking implied. Where a cited publication's compound panel includes it, that fact is noted and no data for it is recorded.

What this pass did: re-retrieved the two labels in full (57-page USPI, 68-page SmPC, extracted and searched locally), re-opened Karoulia 2016, King 2013, Holderfield 2014, Adelmann 2016 and the four PDB entries at source, and pulled the Flaherty 2012 and Robert 2015 abstracts from the Europe PMC record rather than from secondary summaries. Corrections below are applied inline.

1. Resolved identity

INN dabrafenib; marketed as dabrafenib mesylate. Code numbers GSK2118436, GSK2118436A; GSK2118436B = mesylate salt. SB-590885 is a different GSK RAF tool compound and must not be conflated. GSK2366297 (BRAF tool) and GSK2091976 (MEK tool) are also distinct GSK tool compounds and are load-bearing below. Brand TAFINLAR — capsules (50 mg, 75 mg) and tablets for oral suspension (10 mg, paediatric). Originator GlaxoSmithKline. First FDA approval 29 May 2013, BRAF V600E unresectable/metastatic melanoma, applicant GlaxoSmithKline; approved alongside the THxID-BRAF companion assay. Current sponsor Novartis Pharmaceuticals Corporation (2015 GSK/Novartis oncology asset swap). US applications — CORRECTED: capsules NDA 202806; tablets for oral suspension NDA 217514. Both are maintained together (e.g. supplement letters 202806/S-036 and 217514/S-010, 2025). The audit named only 202806. EU — ADDED: MAH Novartis Europharm Limited, Dublin; EU/1/13/865; date of first authorisation 26 August 2013; latest renewal 08 May 2018. Partner MEK inhibitor trametinib (GSK1120212, MEKINIST), same ownership history.

Binding-mode class — DOWNGRADED FROM ASSERTION TO NOMENCLATURE. The audit overrode the task's "Type I, early generation" with a flat "Type I½ (DFG-in / alphaC-out)". Karoulia's own term is alphaC-OUT; "Type I½" is review-tier taxonomy applied inconsistently across the structural literature. Neither label states any binding mode. Record the structural observation, not the taxonomy label. [FLAG: if a binding-mode class is needed for a comparison table, state which review's taxonomy is being used and cite it as review tier.]

2. Binding mode — venue: peer (Karoulia) + PDB; labels are silent

Karoulia Z et al., Cancer Cell 2016;30(3):485-498, doi:10.1016/j.ccell.2016.06.024, PMID 27523909 (verified at PMC5021590) places dabrafenib in the alphaC-OUT class, with this qualifier — quote corrected, the audit had truncated it:

"DAB and AZ-VEM, although alphaC-OUT inhibitors, promote RAF priming and dimerization more potently than VEM and PB and at levels closer to alphaC-IN inhibitors"

and

"DAB and AZ-VEM stabilize the alphaC-helix in an overall OUT position, but R506 in a further IN position compared to PB and VEM, and closer to the position stabilized by alphaC-IN inhibitors"

CORRECTED COMPOUND IDENTITY. The audit wrote "AZ-vemurafenib". No such agent exists. The paper's compound is AZ-VEM, a chimeric tool compound — "an AZ-like core structure attached to the same substituted sulphonamide group from VEM". It is not vemurafenib, not a clinical asset, and not a marketed drug. Any downstream text carrying "AZ-vemurafenib" is wrong.

Profiled panel (verified): AZ-628, TAK-632, LY3009120, GDC-0879, SB-590885, PLX7904 (PB), vemurafenib (VEM), dabrafenib (DAB) — plus the chimeric AZ-VEM used in the structural/priming comparison, which the audit omitted.

R506 (alphaC-helix, dimer-interface-proximal) remains the one published, named residue whose position is load-bearing for dabrafenib. Beyond R506, no residue-by-residue contact map was verified — unknown, not inferred.

Co-crystals (all four verified at RCSB this pass): - 4XV2 — "B-Raf Kinase V600E oncogenic mutant in complex with Dabrafenib", ligand P06, biological assembly a homodimer (C2, A₂), dabrafenib bound in both chains. CORRECTED: the audit recorded this as "unverified, secondary sources only". It is verified, it is the V600E structure, and it is the most directly relevant of the four. - 5CSW — "B-RAF in complex with Dabrafenib", ligand P06; construct carries 15 engineered mutations, E. coli-expressed. - 6V2U — "Crystal structure of the insect cell-expressed WT-BRAF kinase in complex with Dabrafenib", S. frugiperda, 3.78 Å. Wild-type, not V600E. - 5HIE — "BRAF Kinase domain b3aC loop deletion mutant in complex with dabrafenib", T. ni, four chains.

Label silence, verified by retrieval not by search miss: the strings "dimer" and "monomer" appear ZERO times in the full 3/2026 TAFINLAR USPI (57 pp.) and ZERO times in the Tafinlar SmPC (68 pp.). "CRAF" appears once in each. Neither label discloses any structural or binding-mode information.

3. Monomer vs dimer

1. Karoulia 2016 — QUOTE CORRECTED. This is the single most consequential correction in the file.

Audit rendered: "DAB and VEM showed potent inhibition of monomeric BRAF-V600E but 30 to 100-fold less potency..." Source reads:

"PB, DAB and VEM showed potent inhibition of monomeric BRAFV600E but 30 to 100-fold less potency for inhibition of dimeric BRAFV600E"

The audit deleted the first compound. Consequence: the 30–100× dimer potency loss is a property shared across the compared alphaC-OUT compounds in that assay, not a dabrafenib-discriminating result. Dabrafenib's dimer weakness is real and well evidenced; it is not, on this experiment, distinctive. [FLAG: any comparison table built on "dabrafenib loses 30–100× on dimers" must not imply the comparators do not.]

Assay: paired biochemical/cellular assays of monomeric vs enforced-dimeric BRAF V600E across the panel above, plus RAF-priming/RAS-GTP-complex assays and crystallographic alphaC/R506 positioning.

2. Yao Z et al., Cancer Cell 2015;28(3):370-383, doi:10.1016/j.ccell.2015.08.001, PMID 26343582 — MEASUREMENT RE-STATED. What was measured: "More than 30-fold higher concentrations of vemurafenib were required to inhibit ERK driven by p61 V600E dimers than by p61 V600E R509H monomers." This is a cellular ERK-inhibition potency shift for vemurafenib, not a second-protomer binding-affinity ratio. The audit's phrasing ("relative affinity for the first protomer site is ~30-fold higher than for the second once the first is drug-occupied") converts a functional ratio into a biophysical one. No dabrafenib-specific second-site figure was verified. [FLAG: monomer IC50 "~100-300 nM" in SKBR3 not verified this pass — do not quote.]

3. Label-tier biochemistry — verified verbatim, USPI 3/2026 §12.1:

"Dabrafenib is an inhibitor of some mutated forms of BRAF kinases with in vitro IC50 values of 0.65, 0.5, and 1.84 nM for BRAF V600E, BRAF V600K, and BRAF V600D enzymes, respectively. Dabrafenib also inhibits wild-type BRAF and CRAF kinases with IC50 values of 3.2 and 5.0 nM, respectively, and other kinases, such as SIK1, NEK11, and LIMK1 at higher concentrations."

Selectivity window over wild-type RAF: ~4.9× (WT BRAF) to ~7.7× (CRAF). OVERREACH REMOVED: the audit called these "monomeric-enzyme biochemical values". The label discloses no assay format and no oligomeric state. "Monomeric" was the auditor's inference and has been struck.

Not disclosed: any sponsor dimer IC50; any sponsor monomer-vs-dimer selectivity assay; any claim of monomer/dimer equipotency.

4. Behaviour by alteration

Monomeric BRAF V600E — potent, clinically validated. Biochemical IC50 0.65 nM (§12.1). Cellular: King 2013, verbatim — "In cell lines encoding BRAFV600E, dabrafenib inhibited pERK and pMEK in a concentration-dependent manner with IC50 values of 3 nM and 6 nM, respectively" (corrected: pERK 3 nM, pMEK 6 nM — not a "3–6 nM range"). All approvals V600E/V600K-restricted; §2.1 requires confirmed mutation status before dosing.

Dimeric BRAF fusions (KIAA1549::BRAF) — UNKNOWN for dabrafenib, and the gap is now firmer than the audit stated. Sievert AJ et al., PNAS 2013;110(15):5957-5962 established that KIAA1549-BRAF functions as a homodimer, is resistant to PLX4720, and shows CRAF-independent paradoxical MAPK activation — verified, and verified that the tested RAF inhibitors were PLX4720 and a second-generation Plexxikon compound (PLX PB-3), not dabrafenib. Additional check this pass: Jain P et al., Oncotarget 2017;8(49):84697-84713, doi:10.18632/oncotarget.20949 — the other principal BRAF-fusion pharmacology paper — also did not test dabrafenib experimentally. Dabrafenib's paediatric glioma approvals are V600E-restricted. Any statement that dabrafenib paradoxically activates fusion-driven pLGG is an unmeasured class inference.

Class II (constitutive dimers: K601E, L597, G469A, fusions) — not inhibited at clinically achievable exposure (Yao 2015). Clinical record limited to case reports of dabrafenib plus trametinib, all confounded by the MEK inhibitor. CORRECTED: the audit suggested NCI-MATCH sub-protocols H and R might hold class-resolved data. EAY131-H enrolled BRAF V600(E)-mutant tumours only (melanoma, thyroid and colorectal excluded; NSCLC excluded after approval) — it cannot contain class II or III data. Sub-protocol R is trametinib monotherapy — no dabrafenib at all. Neither closes the gap. No prospective dabrafenib-monotherapy class II dataset exists.

Class III (kinase-impaired: D594G, G466V, N581S) — Yao Z et al., Nature 2017;548:234-238, doi:10.1038/nature23291, PMID 28783719. Class 3 mutants bind RAS-GTP more tightly than wild type, amplify ERK signalling through CRAF, and require coexistent RAS activation; the therapeutic conclusion is upstream RAS/RTK inhibition, not RAF inhibition. QUOTE REMOVED: the audit's quotation ("can trigger paradoxical ERK pathway activation by binding to a drug-free protomer") could not be verified as verbatim and is mechanistically inverted — drug binds one protomer and transactivates the drug-free one. [FLAG: re-read Yao 2017 at page level before quoting; and note this study is class-level — dabrafenib is not separately resolved, so it carries the same class-inference limitation the audit correctly applies to fusions.]

RAS-mutant, BRAF wild-type — frank paradoxical activation, CRAF-dependent, demonstrated with dabrafenib itself. King AJ et al., PLOS ONE 2013;8(7):e67583, all three quotes verified verbatim this pass:

"HCT-116 cells showed increased pMEK and pERK following dabrafenib treatment at 100 nM or 300 nM" "Following dabrafenib treatment, MAPK upregulation was not sensitive to ARAF or BRAF knockdown, but was reduced to baseline upon CRAF depletion"

Adelmann 2016 quantifies the induction in HaCaT-HRAS-G12V keratinocytes: peak ERK induction 2.76 ± 0.34-fold, pERK-induction EC80 59 nM (both verified).

What the MEK partner masks. Trametinib does not stop dabrafenib priming or transactivating RAF; it truncates the pathway one node downstream, so the paradox output is suppressed while the mechanism is untouched. Masking is real and large (COMBI-d, below). Mechanism untouched: RAS-associated non-cutaneous malignancies still occur on the combination (SmPC §4.4 names colorectal and pancreatic cancer), and non-cutaneous malignancies in COMBI-AD were 1% (5/435) combination vs <1% (3/432) placebo (SmPC §4.8, verified verbatim).

5. Paradox claims, corrected

5.1 — LABEL — USPI 3/2026 §5.2 "Tumor Promotion in BRAF Wild-Type Tumors" — verdict: supported (as label text)

"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells which are exposed to BRAF inhibitors. Confirm evidence of BRAF V600E or V600K mutation status prior to initiation of TAFINLAR as a single agent or in combination with trametinib."

Retrieved and read in full; heading and text verified verbatim. Evidence basis: the label cites no study, cell line or concentration, and states the finding at class level ("BRAF inhibitors"), not for dabrafenib. The dabrafenib-specific experiment that underpins it is King 2013 — sponsor-authored (GSK) — which the label does not reference. [FLAG: "supported" here means the label makes this claim, verified. It is not a label-tier experimental demonstration for dabrafenib specifically.]

5.2 — LABEL — USPI 3/2026 §1.8 Limitations of Use — verdict: supported — NEW, MISSED BY THE AUDIT

"TAFINLAR is not indicated for treatment of patients with wild-type BRAF solid tumors [see Warnings and Precautions (5.2)]."

A paradox-driven negative indication written into the label and cross-referenced to the tumor-promotion warning. Together with the discontinuation rule below, this is the strongest regulatory operationalisation of the paradox in the dabrafenib record. Its omission from the audit's claim list was a material gap.

5.3 — LABEL — USPI 3/2026 §5.1 (Non-Cutaneous Malignancies) + §2.4 — verdict: DOWNGRADED, supported → partially-supported

"Based on its mechanism of action, TAFINLAR may promote the growth and development of malignancies with activation of RAS through mutation or other mechanisms." §2.4 Table 5, verbatim: "Non-Cutaneous RAS Mutation-positive Malignancies — Permanently discontinue TAFINLAR."

Why downgraded: the label's own wording is "Based on its mechanism of action" — explicit mechanistic inference. The accompanying incidence (non-cutaneous malignancies in 1% of the pooled adult monotherapy and combination populations) is not RAS-genotype-resolved, so the rule and the number are not linked by evidence. This remains a striking regulatory act; it is not an experiment. [FLAG: how many of the 1% were RAS-confirmed is not stated in either label.]

5.4 — LABEL — Tafinlar SmPC §4.4 "Non-cutaneous malignancies" — verdict: supported, qualified

Verified verbatim from the retrieved SmPC, including the careful attribution of chronic myelomonocytic leukaemia and non-cutaneous head-and-neck SCC to "another BRAF inhibitor" — i.e. not dabrafenib — while naming pancreatic adenocarcinoma and bile duct adenocarcinoma on dabrafenib monotherapy and colorectal and pancreatic cancer on dabrafenib + trametinib. This is the strongest dabrafenib-specific clinical paradox evidence in the record and it sits at label tier. QUALIFIED: the SmPC does not state that RAS mutation was confirmed in the individual cases; "RAS-associated" is the regulator's characterisation of the tumour types. [FLAG: do not report these as genotype-confirmed RAS-mutant tumours.]

5.5 — PEER (sponsor-authored) — King 2013, CRAF dependence — verdict: supported

Quotes verified verbatim (above). CRAF dependence directly demonstrated by siRNA epistasis, not asserted. Honest limit retained: the paper does not itself measure RAF dimerisation; it invokes the published dimer model as context. Authors are GSK employees — declared.

5.6 — PEER — King 2013, MEKi abrogates the paradox — verdict: partially-supported (audit's call upheld and strengthened)

"Rats treated with GSK2366297 alone showed a generalized skin crusting with epithelial hyperplasia and hyperkeratosis. Epithelial hyperplasia/hyperkeratosis was not observed in control rats or the combination of GSK2366297 with GSK2091976"

Verified verbatim. The in-vivo skin arm used a 12-day rat repeat-dose study with BRAF and MEK tool compounds, GSK2366297 selected for structural/activity similarity to dabrafenib and faster lesion onset — not dabrafenib + trametinib. The paper's own title flags this. The cellular pERK-abrogation arm is the better-grounded half.

5.7 — PEER — Flaherty 2012, BRF113220 — verdict: partially-supported (call upheld; population corrected)

Abstract verified in full. cuSCC was one of three co-primary endpoints — "The primary end points were the incidence of cutaneous squamous-cell carcinoma, survival free of melanoma progression, and response." Result: "Cutaneous squamous-cell carcinoma was seen in 7% of patients receiving combination 150/2 and in 19% receiving monotherapy (P=0.09)." Authors' own conclusion: "the rate of proliferative skin lesions was nonsignificantly reduced." CORRECTED population: open-label; 247 enrolled overall, 85 in the phase 1 portion, 162 randomised across dabrafenib 150 + trametinib 1, dabrafenib 150 + trametinib 2, and dabrafenib monotherapy — not "162 to combination arms". Correctly read: hypothesis-generating, not effect-establishing. The 7% vs 19% figure is widely quoted without P=0.09.

5.8 — PEER — Long 2014, COMBI-d — verdict: supported

"fewer patients in the dabrafenib–trametinib group than in the dabrafenib-only group had incident cutaneous squamous-cell carcinoma (2% vs. 9%) or cutaneous hyperkeratoses (3% vs. 32%)" — verified. n=423, double-blind, placebo-controlled, identical dabrafenib dose in both arms: the only design in the record where the MEK inhibitor is the sole variable. The hyperkeratosis delta is the cleanest clinical paradox pharmacodynamic readout available. DISCREPANCY THE AUDIT DID NOT FLAG: NEJM reports 9% cuSCC in the monotherapy arm; the SmPC reports 10% (22/211) for dabrafenib monotherapy in the same study (MEK115306). Both verified; they do not agree. [FLAG: state which source is being quoted; do not present the SmPC figure as "corroborating" the NEJM figure.]

5.9 — VENUE RETAGGED: conference → PEER — Robert 2015, COMBI-v — verdict: partially-supported

Verified: "Cutaneous squamous-cell carcinoma and keratoacanthoma occurred in 1% of patients in the combination-therapy group and 18% of those in the vemurafenib group" — NEJM 2015;372:30-39, n=704, open-label; SmPC cites the same study as MEK116513 (cuSCC 18%, 63/349 on vemurafenib). The audit tagged this "conference" on the ESMO 2014 presentation. The figures it quotes are peer-reviewed and label-corroborated; tagging them provisional misrepresents the record in the other direction. Substantive limitation stands: confounded for the paradox question — the comparator is a different BRAF inhibitor as monotherapy, so the delta conflates MEKi masking with an inter-agent difference. COMBI-d is the correct comparison; COMBI-v is a commercial one.

5.10 — PEER — Karoulia 2016 — verdict: supported, quotes corrected

See §2 and §3 for the two corrected quotes and the AZ-VEM identity fix. The finding that dabrafenib promotes RAF priming and dimerisation more potently than VEM and PB is verified and is genuinely counter-intuitive. [FLAG: priming/dimerisation potency and clinical cuSCC incidence are different axes — do not present this as a direct rebuttal of a toxicity-incidence claim.] Bollag is Plexxikon-affiliated — declared.

5.11 — PEER — Yao 2015 — verdict: DOWNGRADED, supported → partially-supported

Dimer resistance stands. The "~30-fold second-site affinity" framing does not (see §3). Vemurafenib-measured; no verified dabrafenib figure.

5.12 — PEER — Yao 2017 — verdict: DOWNGRADED, supported → partially-supported

Class-level study; dabrafenib not separately resolved; audit's quotation unverified and inverted (see §4). Applying it to dabrafenib is the same class inference the audit refuses for fusions — apply one standard.

5.13 — PEER — Adelmann 2016, paradox index — verdict: partially-supported (call upheld; numbers all verified)

Verified from source: paradox index = pERK-induction EC80 ÷ IC80 against A375.

	A375 IC80	HaCaT-HRAS-G12V pERK EC80	Paradox index	Peak pERK induction
Vemurafenib	380 nM	2,100 nM	5.5	6.86 ± 1.27×
Dabrafenib	5.9 nM	59 nM	10	2.76 ± 0.34×
Encorafenib (LGX818)	2.6 nM	130 nM	50	4.08 ± 0.16×

Clinical rates quoted by the paper, verbatim: "Monotherapy with vemurafenib, dabrafenib, and encorafenib induces neoplasia, most often cutaneous squamous cell carcinoma (cuSCC), at rates of approximately 22%, 6%, and 3.7%, respectively" — averaged across phase I–III trials. Where claim and evidence come apart, unchanged and correct: (i) a three-point correlation; (ii) pooled cross-trial rates with differing dermatologic surveillance; (iii) one engineered keratinocyte line standing in for all paradox-competent tissue; (iv) the index measures pERK output only and is silent on dimer promotion, where Karoulia ranks dabrafenib worse than vemurafenib. NOTE: the published panel includes a fourth agent that is outside this file's scope; no data for it is recorded here.

5.14 — REVIEW — Holderfield, Nagel & Stuart 2014 — verdict: DOWNGRADED, partially-supported → asserted-only

Quote corrected — the audit deleted the hedge:

"It has been speculated that the lower incidence of cutaneous lesions seen with dabrafenib is owing to higher potency against BRAFV600E compared with WT BRAF and CRAF, whereas vemurafenib may be relatively equipotent"

Br J Cancer 2014;111:640-645, doi:10.1038/bjc.2014.139, PMID 24642617. Full text is open at PMC4134487 — the audit's "publisher auth wall" was an unattempted retrieval, and opening it reversed the finding. Affiliations corrected: M Holderfield — UCSF Helen Diller Family Comprehensive Cancer Center. T E Nagel and D D Stuart — Novartis Institutes for Biomedical Research, Emeryville. The audit described all three as Novartis-affiliated; the first author was not. No new experiment. As a speculation in a secondary source it is asserted-only, not partially-supported. The three primary-source tensions remain valid reasons not to propagate it: the label's own ~4.9×/7.7× selectivity window is modest, not "highly selective"; Karoulia shows dabrafenib promotes priming/dimerisation more potently than vemurafenib; and the cuSCC difference it explains is cross-trial and unadjusted for surveillance intensity. [FLAG: the audit built its critique on a quote it had altered — the review had already marked this as speculation.]

5.15 — SPLIT, was a single "supported" peer claim — "paradox inducer, not paradox breaker"

Positive half — verdict: supported (peer). Dabrafenib is characterised in the primary RAF-dimer literature as a paradox inducer and is markedly less potent against dimeric BRAF V600E (Karoulia 2016; Yao 2015). Where it appears in paradox-breaker papers it is the comparator that induces the paradox. Both labels carry the tumour-promotion warning. Negative half — verdict: DOWNGRADED to asserted-only; venue tag "peer" withdrawn. The proposition "no label, peer, conference or company source anywhere makes a dimer-selectivity or paradox-breaking claim for dabrafenib" rests entirely on the auditor's own unsuccessful searching. A negative search result is not a source and cannot be peer tier. [FLAG: usable as "none located in this pass", never as "none exists".] For CI purposes the defensible statement is unchanged: any deck positioning dabrafenib as paradox-sparing in absolute terms is unsupported; the only defensible relative statement is "lower cuSCC rate than vemurafenib in cross-trial comparison, with a contested mechanistic explanation."

5.16 — PEER — Sievert 2013, BRAF fusions — verdict: asserted-only for dabrafenib (upheld and strengthened)

PNAS 2013;110(15):5957-5962, doi:10.1073/pnas.1219232110, PMID 23533272. Verified: KIAA1549-BRAF functions as a homodimer, is resistant to PLX4720, and shows CRAF-independent paradoxical MAPK activation; a second-generation Plexxikon compound (PLX PB-3) was the other agent tested. Dabrafenib was not tested. Strengthened this pass: Jain P et al., Oncotarget 2017;8(49):84697-84713 — the other principal BRAF-fusion pharmacology paper — also did not test dabrafenib. The proposition "dabrafenib paradoxically activates BRAF-fusion-driven tumours" remains an unmeasured class inference. Recording it as asserted-only is the finding.

6. Clinical paradox signal — all figures re-verified against retrieved label text

US label (TAFINLAR USPI rev. 3/2026, §5.1) — verified verbatim: - Dabrafenib monotherapy, pooled adult: cuSCC 11%, keratoacanthoma 4%; basal cell carcinoma 4%, new primary melanoma 1%. Denominators not stated in §5.1; cross-referenced to §6.1. - Dabrafenib + trametinib, pooled adult: cuSCC incl. keratoacanthoma 2%; BCC 3%; new primary melanoma <1%. - Paediatric combination: new primary melanoma <1%. - Non-cutaneous malignancies 1% across pooled adult monotherapy and combination. - Risk management: dermatologic evaluation before initiation, every 2 months on therapy, and for up to 6 months after discontinuation. §2.4: permanent discontinuation for RAS mutation-positive non-cutaneous malignancies. §1.8: not indicated in wild-type BRAF solid tumours.

EU label (Tafinlar SmPC §4.4, §4.8) — verified verbatim, with denominators: - cuSCC on dabrafenib monotherapy, MEK115306: 10% (22/211); on vemurafenib monotherapy, MEK116513: 18% (63/349) — comparator, not dabrafenib. - cuSCC on dabrafenib + trametinib across pivotal melanoma and NSCLC studies: 2%; median time to onset 18–31 weeks. - Time-to-onset within MEK115306: 223 days (range 56–510) combination vs 60 days (range 9–653) monotherapy. - Monotherapy onset profile: "approximately 70% of the events occurred within the first 12 weeks of treatment with a median time to onset of 8 weeks." - COMBI-AD (BRF115532): cuSCC 1% (6/435) combination vs 1% (5/432) placebo at primary analysis; during long-term (up to 10 years) off-treatment follow-up, 2 additional cuSCC in each arm. - Non-cutaneous malignancies: 1% (6/586) in the integrated dabrafenib monotherapy safety population; <1% in pivotal combination studies; COMBI-AD 1% (5/435) combination vs <1% (3/432) placebo. - Named RAS-associated malignancies on dabrafenib monotherapy: pancreatic adenocarcinoma, bile duct adenocarcinoma. On dabrafenib + trametinib: colorectal cancer, pancreatic cancer. CMML and non-cutaneous head-and-neck SCC are explicitly assigned to "another BRAF inhibitor". - EU surveillance mandated beyond skin: head-and-neck examination (baseline; every 3 months on treatment), chest/abdomen CT at baseline and every 6 months, anal and pelvic examinations before and at end of treatment, blood counts as indicated, and 6 months' post-discontinuation monitoring for non-cutaneous malignancy. The weight of this regimen is itself a measure of how seriously the regulator takes the signal.

Randomised trial tier: - BRF113220 (Flaherty 2012): cuSCC 7% (dab 150 + tram 2) vs 19% (dab mono), P=0.09, co-primary endpoint, open-label, 247 enrolled / 162 randomised across three arms. Authors: "nonsignificantly reduced." - COMBI-d (Long 2014): cuSCC 2% vs 9%; hyperkeratosis 3% vs 32%; n=423; double-blind, placebo-controlled, same dabrafenib dose both arms. The definitive masking evidence. - COMBI-v (Robert 2015): cuSCC/KA 1% vs 18% (vemurafenib); n=704; confounded — different BRAF inhibitor in the control arm.

Consistency verdict (unchanged, and it survives the adversarial pass). The clinical readouts are fully consistent with paradoxical activation; there is no clinical signal against it. Dabrafenib monotherapy produces the canonical triad — cuSCC 10–11%, keratoacanthoma 4%, new primary melanoma 1%, hyperkeratosis ~32% — with early onset (median 8 weeks, ~70% within 12 weeks) matching a pharmacologically driven proliferative effect rather than sporadic skin cancer. Adding trametinib reduces cuSCC ~4–5× and delays onset ~3.7×, masking the cutaneous readout. It does not abolish the RAS-driven tumour-promotion signal: RAS-associated non-cutaneous malignancies are still reported on the combination at label tier, and the discontinuation rule and the wild-type-BRAF non-indication apply on either regimen.

Real-world context: Sauer N, Giedziun P, Calik J, Wiela-Hojenska A. Cutaneous Adverse Drug Reactions Associated With BRAF and MEK Inhibitors: A Real-World Analysis of WHO Pharmacovigilance Data. Clin Pharmacol Ther 2026;120:274-285, doi:10.1002/cpt.70326 — abstract level only; no dabrafenib-specific disproportionality statistics extracted or quoted. [FLAG: full text not read; do not cite any number from it.]

7. Unknowns and honest absences

Company tier is NOT empty — the audit's finding is corrected. A company-tier source exists: the GSK press release of 29 May 2013 announcing FDA approval, which restates the §5.2 paradox sentence verbatim and additionally reports BREAK-3 comparator figures (cuSCC in 7% of dabrafenib-treated patients vs none on chemotherapy; new primary malignant melanoma 2% (3/187) vs none on chemotherapy). Awareness tier only — it establishes nothing. But the audit's hedge ("nothing distinct from a label restatement") survives, and the tier should be recorded as populated with label-restatement material, not empty. Historical GSK 2012–2015 investor decks and Novartis oncology-day material were not exhaustively searched. [FLAG: the 7% / 2% (3/187) BREAK-3 figures are company tier and must not be used as evidence; retrieve Hauschild 2012 or the FDA medical review to establish them.]
No sponsor-disclosed dabrafenib dimer IC50 or monomer:dimer selectivity ratio has ever been published. The only quantitative figure (30–100× less potent against dimeric than monomeric BRAF V600E) is from an independent academic lab — and, corrected, applies to PB, DAB and VEM in that experiment.
No publication tests dabrafenib against KIAA1549::BRAF or any dimeric BRAF fusion. Confirmed across Sievert 2013 (PLX4720, PLX PB-3) and Jain 2017 (PLX8394-class agent, PLX4720, vemurafenib, MEK inhibitors). Genuinely unmeasured for dabrafenib.
No prospective dabrafenib-monotherapy dataset in class II or class III BRAF mutants, and no NCI-MATCH arm can supply one — H is V600-restricted, R is trametinib-only. Every class II clinical datapoint located is dabrafenib plus trametinib.
Residue-level contact map beyond R506 is not established. The four PDB entries are now verified (4XV2, 5CSW, 6V2U, 5HIE) but their primary papers were not read.
USPI §6.1 was not extracted — the denominators behind the pooled monotherapy percentages (11% / 4% / 4% / 1%) remain unread. This is a retrievable gap in the same PDF, not an absence in the record. [FLAG: extract §6.1 before publishing any US denominator.] The EU integrated monotherapy population, 586, is the only verified denominator of that kind.
RAS-genotype confirmation for the 1% non-cutaneous malignancies is not stated in either label, in either jurisdiction.
BREAK-3 (Hauschild 2012) and Falchook 2012 cuSCC rates remain unverified. Secondary sources conflict (6% vs 10%); a company release gives 7%. Do not quote a BREAK-3 cuSCC number from this file. Use the SmPC's 10% (22/211) for MEK115306 instead — noting it disagrees with NEJM's 9% for the same arm.
Yao 2015 and Yao 2017 were verified through aggregated full-text retrieval, not page-level reading. The quotations retained are the ones that matched; the two the audit used and this pass could not confirm have been struck. [FLAG: page-level re-read required before external use.]
No conference-tier material was located that makes a paradox claim about dabrafenib beyond what was subsequently published peer-reviewed. Recorded as "none located", not "none exists".
No head-to-head clinical trial of dabrafenib against any other BRAF inhibitor as monotherapy exists. Every cross-agent cuSCC comparison in the record is cross-trial, unadjusted for surveillance intensity, and in COMBI-v further confounded by the MEK inhibitor being present in only one arm.

8. Two rules this pass would add for the next dossier

A retrieval you did not attempt is not an unknown. Two of this audit's recorded gaps — the Br J Cancer review and PDB 4XV2 — closed on the first try, and one of them reversed a verdict. Before writing "not disclosed", the retrieval must have been attempted and failed.
Quote integrity is venue integrity. Two of the file's three most consequential errors were deletions inside quotation marks: a dropped compound in the Karoulia dimer sentence and a dropped "It has been speculated that" in the Holderfield sentence. Both changed what the source claimed. Verbatim means verbatim, including the parts that weaken the point. »

Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.

Verification confidence: high

Encorafenib (BRAFTOVI) — Paradox-Claim Audit, Adversarially Re-Verified

Data cut: 1 September 2026. All retrievals executed on or before this date.

Scope exclusion (binding): Plixorafenib / FORE8394 / PLX8394 is out of scope. Where source papers used it as a comparator arm, only the encorafenib arm is reported. No comparison against it is offered or implied.

Verdict axes. This corrected file separates two things the original audit collapsed into one field: - Evidence verdict — does the cited evidence support the claim as stated? - Bearing — which way does it cut on a "paradox-sparing" narrative?

An absence can be supported as a finding while contradicting a narrative. The original file recorded several of these as "contradicted", which is a category error.

1. Resolved identity

Field	Value	Tier
INN	encorafenib	label
Research codes	LGX818, LGX-818, NVP-LGX818, NVP-LGX-818-NXA, NVP-LGX818-NXA	ChEMBL CHEMBL3301612
CAS	1269440-17-6	registry
Trade name	BRAFTOVI, hard capsules 50 mg / 75 mg	label
ATC	L01EC03	EU SmPC §5.1
First approval	2018	ChEMBL / regulatory
Originator	Novartis (Genomics Institute of the Novartis Research Foundation)	conference / company
Divestment	Array BioPharma acquired worldwide rights from Novartis Pharma AG, announced January 2015, $85M upfront	company
Current US sponsor	Pfizer (Array acquired 2019); NDA 210496	label
EU MAH	[FLAG: see below]	—

CORRECTED — fabricated code numbers struck. The prior audit listed "ARRY-085 / ARRY-424704 (Array codes appearing in some filings)". Neither appears in the ChEMBL synonym set, and Array's own release is titled "Array Announces Agreement To Acquire ENCORAFENIB (LGX818)" with no ARRY designation. Both codes are removed. This is the identity error the brief warned about: newer INNs and their sponsor codes are frequently cross-contaminated, and an unverifiable code number propagates into every downstream search.

[FLAG: EU Marketing Authorisation Holder. The prior audit asserted "PIERRE FABRE MEDICAMENT (Lavaur, France) per the Braftovi EPAR Annex I." The EMA EPAR PDF would not text-extract at this data cut (returned undecoded FlateDecode streams, 890 KB binary). The document I did retrieve and read in full — the UK emc Braftovi 75 mg SmPC — names Pierre Fabre Limited. A human must open the EMA EPAR Annex I and confirm which entity is named there. Do not cite the Lavaur address until this is done.]

2. Binding mode — and the class argument that must be retired

What is established (peer): encorafenib is a type I½ inhibitor stabilising a DFG-in / αC-helix-out conformation. Verbatim, Marini E et al., Molecules 2022;27(23):8513, PMID 36500607: "Both BI 882370 and encorafenib are type I½ RAF kinase inhibitors, and stabilize a DFGin and αC-helixout conformation." The EU SmPC independently describes it as "a potent and highly selective ATP-competitive small molecule RAF kinase inhibitor" — ATP-competitive, consistent with type I/I½, not type II.

No co-crystal exists. Now demonstrated, not asserted. Marini verbatim: "While no experimental structure of encorafenib in BRAF is available, two X-ray structures of BI 882370 in BRAF can be found in the PDB."

CORRECTED — the RCSB search was asserted in the prior audit; it has now actually been run. Full-text query encorafenib against the RCSB search API returns total_count = 1: entry 9BP8, "Crystal structure of BRAF kinase domain with PF-07799933", X-ray, 1.73 Å, UniProt P15056, ligand A1AQY, deposited 2024-05-07, released 2025-05-14. Encorafenib is not the bound ligand. The no-co-crystal conclusion survives and is now evidenced — but the prior audit's phrasing implied a null result, and the search is not null.

CORRECTED — the pose is docked, not homology-modelled. Marini docked encorafenib onto the experimental BRAF–dabrafenib co-crystal PDB 5CSW (2.66 Å) after self-docking validation of dabrafenib. The two BI 882370 surrogates were rejected for different reasons the prior audit merged: 5CSX rejected as a DFG-out crystallisation artefact; 6UUO (3.29 Å) rejected for a disordered P-loop and activation loop.

CORRECTED — contact residues, verbatim. "hydrogen bonds between residues Asp594 and Phe595 (DFG segment), Lys483 and the sulfonylamide moiety" and "a typical 1–3 hydrogen bond with the backbone of Cys532 in the hinge region." The prior audit's "pyrimidinyl interaction with the Cys532 backbone nitrogen" is not in the retrieved text.

OVERREACH REMOVED — the class-exclusivity claim

The prior audit asserted that αC-OUT type I½ binding "is the conformational class that is mechanistically ASSOCIATED with RAF transactivation and paradox." The definite article is refuted.

Tkacik E, Jang DM, Boxer K, Ha BH, Eck MJ. In vitro reconstitutions suggest a general model for paradoxical activation of ARAF, BRAF, and CRAF by diverse RAF inhibitor types that does not rely on negative allostery. eLife reviewed preprint RP110344, v1, 23 March 2026. doi 10.7554/eLife.110344.1 — type I AND type II inhibitors induce paradoxical activation of all three RAF isoforms; the inhibition phase for type II inhibitors is suggestive of positive cooperativity, in conflict with the prevailing negative-allostery model. See also eLife 95481, "Allosteric coupling asymmetry mediates paradoxical activation of BRAF by type II inhibitors."

Consequence, stated plainly: paradoxical activation is not a property of the αC-OUT / type I½ class alone. Encorafenib's binding class cannot be used to argue paradox in either direction. Encorafenib's paradox must be established by direct experiment on encorafenib — which it is, below. Retire the structural line of argument rather than caveating it; a single docking run onto a different drug's crystal cannot carry a mechanistic conclusion.

[FLAG: Tkacik does not test encorafenib — it names it once, in the Introduction only. It is used here solely to refute a class-exclusivity generalisation, never as evidence about encorafenib itself.]

3. Monomer vs dimer

MONOMER-SELECTIVE by class assignment. The quantitative comparison is NOT DISCLOSED.

Cell-free enzyme IC50s — label tier, both jurisdictions, verbatim.

FDA USPI §12.1 (rev. 2/2026): "Encorafenib is a kinase inhibitor that targets BRAF V600E, as well as wild-type BRAF and CRAF in in vitro cell-free assays with IC50 values of 0.35, 0.47, and 0.3 nM, respectively."

EU SmPC §5.1: "The half maximal inhibitory concentration (IC50) of encorafenib against BRAF V600E, BRAF and CRAF enzymes was determined to be 0.35, 0.47 and 0.30 nM, respectively."

Kinase	IC50	Venue
BRAF V600E	0.35 nM	label (FDA §12.1, EU §5.1)
Wild-type BRAF	0.47 nM	label
CRAF	0.30 nM	label

ADDED — a first-order label-tier fact the prior audit reported but did not read

Encorafenib has no enzymatic selectivity for V600E over wild-type RAF. CRAF is inhibited more potently than the mutant target (0.30 vs 0.35 nM), and wild-type BRAF is near-equipotent (0.47 nM).

This requires no inference, no structure, and no modelling. It is the direct biochemical substrate for paradoxical activation, it is stated at label tier in both jurisdictions, and it sits in the same EU SmPC sentence-pair that calls the molecule "potent and highly selective." The prior audit transcribed all three numbers and drew no consequence from them — while spending several paragraphs on a docked pose. This is the strongest, cleanest, highest-tier fact in the file and it was passed over.

ADDED — label-tier off-target binding. FDA §12.1 verbatim: "Encorafenib was also able to bind to other kinases in vitro including JNK1, JNK2, JNK3, LIMK1, LIMK2, MEK4, and STK36 and reduce ligand binding to these kinases at clinically achievable concentrations (≤0.9 µM)." Absent from the prior audit; bears on any "highly selective" framing.

Qualitative class assignment (peer, but re-cited not generated). Pickles OJ et al., Oncotarget 2020;11(34):3188–3197, PMID 32922659, Introduction, cited to references [4]–[6]: "Encorafenib is a group 1 BRAF inhibitor that selectively inhibits active BRAF monomers. However, in RAF dimers, binding of the inhibitor to one RAF protomer activates the other, a process known as negative co-operativity." This is an assertion re-cited from prior literature, not a result of that paper.

Functional dimer-occupancy behaviour (peer, primary). Cotto-Rios XM et al., Nat Commun 2020;11:4370, PMC7462985 — Supplementary Fig. 19: "Encorafenib treatment for 1 h inhibited p-ERK, but the activation is recovered after wash-out for another hour due to half-occupied active dimers." Condition, now disclosed: 1 µM encorafenib, 1 h, wash 3× PBS, 1 h recovery. Single high concentration, tool condition, not a dose–response.

NOT DISCLOSED: no published side-by-side IC50 of encorafenib against monomeric versus dimeric BRAF V600E; no measured cooperativity coefficient for encorafenib's own second-site binding; no dimer-selectivity ratio.

CORRECTED citation. Clayton J, Romany A, Matenoglou E, Gavathiotis E, Poulikakos PI, Shen J. Mechanism of dimer selectivity and binding cooperativity of BRAF inhibitors. eLife 2025;13:RP95334, PMID 39945510 — published, not a preprint as the prior audit stated. Substance confirmed: "encorafenib" appears exactly once, in the Introduction ("the current FDA-approved inhibitors, Vermurafenib, Dabrafenib, and Encorafenib, inhibit the monomeric BRAFV600E"), with no IC50, no cooperativity coefficient, no dimer-selectivity value. Only PHI1 has a full cooperativity analysis in that paper.

4. Behaviour by alteration

Monomeric BRAF V600E (on-target). Potent. Enzyme IC50 0.35 nM (label). Cellular IC50 0.0028–0.0382 µM across BRAF-V600E CRC and melanoma lines — A375 0.0066, G361 0.0382, LS411N 0.0028, Colo 201 0.0205, RKO 0.0269 µM (Pickles 2020, peer, verified). pERK-suppression EC50 3 nM and antiproliferative EC50 4 nM in A375 [FLAG: attributed to Stuart AACR 2012 abstract 3790, full text confirmed non-retrievable — HTTP 403 — at this data cut. Not present in indexed excerpts. Unverified at primary level.]

CORRECTED jurisdiction. "Encorafenib suppresses the RAF/MEK/ERK pathway in tumour cells expressing several mutated forms of BRAF kinase (V600E, D and K)" is EU SmPC §5.1, not the FDA label. FDA §12.1 names only V600E, wild-type BRAF and CRAF.

Dimeric RAF / BRAF fusions — UNKNOWN. No encorafenib data in any BRAF-fusion (KIAA1549-BRAF or equivalent) dimer-driven model identified in any venue at this data cut. What exists is dimer-context failure by proxy: half-occupied active dimers with pERK recovery within 1 h of washout at 1 µM (Cotto-Rios, peer); and time-dependent pERK reactivation at 3/24/48 h in BRAF-V600E CRC lines (Pickles, peer). The label's own §12.1 concedes EGFR-mediated MAPK reactivation as the CRC resistance mechanism requiring cetuximab.

Class II BRAF (dimer-forming, RAS-independent) — PARADOXICAL ACTIVATION DEMONSTRATED. Cotto-Rios 2020, Figure 7 legend, verbatim: "Notably, Encorafenib pre-treatment promotes paradoxical activation in these cells." Panel includes H2087 (BRAF L597V, class II).

Class III BRAF (kinase-impaired, RAS-dependent) — PARADOXICAL ACTIVATION DEMONSTRATED. Same experiment; panel includes H1666 (BRAF G466V, class III).

RAS-mutant — PARADOXICAL ACTIVATION DEMONSTRATED, in vitro and in vivo.

In vitro, NRAS-mutant melanoma: SKMEL-30 and SKMEL-2 in the same Cotto-Rios panel. [FLAG: genotype assignment discrepancy. The prior audit records SKMEL-30 = NRAS Q61K and SKMEL-2 = NRAS Q61R. Extraction of the Figure 7 legend returns the reverse: SKMEL-30 (BRAF-WT/NRAS Q61R) and SKMEL-2 (BRAF-WT/NRAS Q61K). Standard reference/CCLE genotypes are SK-MEL-2 = Q61R and SK-MEL-30 = Q61K, matching the prior audit and not the extracted legend. Both lines are NRAS Q61 mutants either way, so the paradox conclusion is unaffected — but a human must check the published figure before either assignment is quoted.]
In vitro, HRAS-mutant keratinocyte: a measurable pERK-induction EC80 in HaCaT cells stably expressing HRAS G12V (Adelmann CH et al., Oncotarget 2016;7(21):30453–60, PMID 27028853, peer). Readout at 15 min primary, with a 72-h time course peaking at 6 h. The existence of a finite EC80 is itself the demonstration of paradoxical activation.
In vivo, BRAF-wild-type tissue: "The experimental Type 1 RAF inhibitor, SB590885, and/or encorafenib (a RAF inhibitor used clinically) increased ERK1/2 phosphorylation in cardiomyocytes, and promoted hypertrophy, consistent with a 'RAF paradox' effect, with both promoting cardiac hypertrophy in mouse hearts in vivo." Biochem J. 2022;479(3):401–424, PMID 35147166 (peer, mouse in vivo). [FLAG: the "and/or" construction means figure-level confirmation of which panels used encorafenib specifically, versus SB590885, is still required. Also: the prior audit's "who" field read "Academic cardiac-signalling group" — not a named source, which fails the audit's own rule 2. Retrieve the author list.]

BRAF wild-type, no RAS lesion. "LGX818 was inactive against BRAF wild-type tumors at doses up to 300 mg/kg bid" (Stuart, AACR 2012 abstract 3790, conference — this sentence is verified in indexed abstract text). The EU SmPC's "Encorafenib does not inhibit RAF/MEK/ERK signalling in cells expressing wild-type BRAF" derives from this nonclinical package.

CAUTION, unchanged and correct: a growth-inhibition screen cannot detect paradoxical activation. Absence of antiproliferative effect is not absence of paradox.

NET. Encorafenib is a monomer-selective, ATP-competitive, αC-OUT type I½ RAF inhibitor with no enzymatic selectivity for V600E over wild-type BRAF or CRAF at label tier. Paradoxical activation is demonstrated for encorafenib specifically across class II BRAF, class III BRAF, NRAS-mutant and HRAS-mutant contexts, and in vivo in wild-type mouse myocardium. Its binding class neither predicts nor excludes this; the direct experiments do.

5. Paradox claims — corrected register

5.1 — LABEL TIER

[1] FDA §5.2, verbatim, re-retrieved and confirmed word-for-word

"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells, which are exposed to BRAF inhibitors. Confirm evidence of BRAF V600E or V600K mutation prior to initiating BRAFTOVI."

Who: FDA / Pfizer, NDA 210496. Source: BRAFTOVI USPI §5.2 "Tumor Promotion in BRAF Wild-Type Tumors", rev. 2/2026 (labeling.pfizer.com id=12990). Venue: label. Evidence basis: the label cites no specific study — it asserts unspecified "in vitro experiments" as a class finding and applies it to encorafenib. Encorafenib-specific grounding comes from outside the label: Cotto-Rios 2020, Adelmann 2016, Biochem J 2022. Evidence verdict: SUPPORTED (claim as stated is verbatim accurate; the underlying assertion is a class statement corroborated independently for this molecule). Bearing: decisively against any paradox-sparing narrative.

[2] FDA §5.1 / EU §4.4 — RAS-mutant malignancy restriction, verbatim

FDA §5.1, Noncutaneous Malignancies: "Based on its mechanism of action, BRAFTOVI may promote malignancies associated with activation of RAS through mutation or other mechanisms... Monitor patients receiving BRAFTOVI for signs and symptoms of noncutaneous malignancies. Discontinue BRAFTOVI for RAS mutation-positive noncutaneous malignancies." EU §4.4, Non-cutaneous malignancies: "Based on its mechanism of action, encorafenib may promote malignancies associated with activation of RAS through mutation or other mechanisms... It should be considered to permanently discontinue encorafenib in patients who develop RAS mutation-positive non-cutaneous malignancies."

Venue: label, both jurisdictions. Evidence basis: mechanistic inference stated by the regulator; no trial or experiment cited in either label. Corroborated post-approval by Cotto-Rios 2020 (NRAS Q61 lines). Evidence verdict: SUPPORTED. Bearing: a regulator-mandated behavioural restriction in both jurisdictions that would be incoherent for a molecule that did not transactivate RAF in a RAS-mutant background.

[3] EU SmPC §5.1 — "does not inhibit RAF/MEK/ERK signalling in cells expressing wild-type BRAF" Venue: label. Source: Braftovi SmPC §5.1, Mechanism of action (retrieved and read in full). Evidence basis: traceable to the Novartis nonclinical package (Stuart AACR 2012, conference). This is a statement about failure to INHIBIT, generated from antiproliferative and xenograft growth screens. It is not, and was never generated to be, a statement about failure to ACTIVATE. Evidence verdict: PARTIALLY-SUPPORTED. Bearing: This sentence is the single most likely source of a mistaken "encorafenib is paradox-sparing" inference. Two facts sit against it inside the same document: (a) the same §5.1 paragraph gives wild-type BRAF IC50 0.47 nM and CRAF 0.30 nM — encorafenib demonstrably engages wild-type RAF; (b) §4.4, three sections earlier, carries the RAS-mutant malignancy discontinuation instruction.

[4] The EU label carries no paradox statement — ABSENCE, partially re-verified Claim: the EU Product Information contains no statement of paradoxical MAPK activation and no equivalent to FDA §5.2. Venue: label (absence documented by reading the label). What I verified: I retrieved and read the UK emc Braftovi 75 mg SmPC in full. "paradox", "paradoxical" and "dimer" do not appear anywhere in it. §4.4 carries cutaneous and non-cutaneous malignancy warnings including the RAS discontinuation instruction, with no paradox language. §5.1 carries the IC50s, the >30 h dissociation half-life and the wild-type sentence, with no paradox language. [FLAG: the prior audit's precise hit counts could not be reproduced. It claimed "'paradox' = 0, 'dimer' = 0, 'residence time' = 0, 'dissociation' = 1 (§5.1), 'wild-type' = 2 (§4.2, §5.1)" across a "63-page EU Product Information (Annexes I–IIIB)." The EMA EPAR PDF would not text-extract at this data cut. My verification is on a different document (UK emc, Annex I equivalent) and does not cover Annexes II–IIIB or the page count. A human must re-run the term search against the EMA EPAR PDF itself before any per-term count is published.] CORRECTED — evidence verdict: SUPPORTED as an absence, on the Annex I / SmPC portion. (The prior audit recorded "contradicted", conflating the evidence axis with the narrative axis.) Bearing: materially exploitable, and the practical warning is unchanged — anyone citing "the label" on encorafenib and paradox must specify WHICH label. Citing the EU SmPC alone yields a false negative.

[5] EU SmPC §5.1 — dissociation half-life

"The encorafenib dissociation half-life was >30 hours and resulted in prolonged pERK inhibition."

Venue: label (EU only). CONFIRMED verbatim. CORRECTED — the FDA label omits it entirely. Re-retrieval of USPI §12.1 confirms: no dissociation half-life, no residence time, anywhere in the FDA prescribing information. The datum exists at label tier in the EU only. ADDED — the qualifier the prior audit dropped. Koelblinger 2018 verbatim: "a substantially increased dissociation half-life (T1/2-diss) from V600E-mutant BRAF of more than 30 h, as compared with 2 and 0.5 h reported for dabrafenib and vemurafenib, respectively" [refs 7, 10]. The residence time is measured on the MUTANT target. It carries no information about dwell time on wild-type BRAF or CRAF — which is where paradox happens. This voids most residence-time-to-paradox reasoning in either direction. Evidence verdict: SUPPORTED (the label says exactly this). Bearing: neutral. The regulator attaches the long residence time to prolonged pERK inhibition in the target cell, and no paradox consequence.

5.2 — PEER TIER, PRIMARY EXPERIMENT

[6] Cotto-Rios XM et al. — paradoxical activation in class II, class III and NRAS-mutant cells Who: Cotto-Rios XM, Agianian B, Gitego N, Zacharioudakis E, Giricz O, Wu Y, Zou Y, Verma A, Poulikakos PI, Gavathiotis E (Einstein / Mount Sinai). Source: Inhibitors of BRAF dimers using an allosteric site. Nat Commun 2020;11:4370, doi 10.1038/s41467-020-18123-2, PMC7462985. Venue: peer, primary. Evidence: direct experiment. Panel — H1666 (BRAF G466V, class III), H2087 (BRAF L597V, class II), SKMEL-30 and SKMEL-2 (both BRAF-WT / NRAS Q61, see genotype flag above). pERK immunoblot with encorafenib pre-treatment and washout. CORRECTED strength disclosure: the paradox sentence sits in the Figure 7 legend; the washout observation in Supplementary Fig. 19; the tool condition is 1 µM encorafenib for 1 h — a single high concentration, not a dose–response. Evidence verdict: SUPPORTED, with the above qualifications. It demonstrates that paradoxical activation occurs; it does not establish its threshold, dose-dependence or magnitude. Bearing: the strongest single primary evidence in the record against any dimer-sparing or paradox-breaking characterisation of encorafenib.

[7] Adelmann CH et al. — paradox index 50 Who: Adelmann CH, Ching G, Du L, Saporito RC, Bansal V, Pence LJ, Liang R, Lee W, Tsai KY (MD Anderson). Source: Comparative profiles of BRAF inhibitors: the paradox index as a predictor of clinical toxicity. Oncotarget 2016;7(21):30453–60, PMID 27028853. Venue: peer, primary. Definition, verbatim: "the paradox index, an in vitro surrogate of a therapeutic index calculated by dividing each pERK induction EC80 by the IC80 against A375." Values confirmed: encorafenib 50, dabrafenib 10, vemurafenib 5.5. Assay, corrected and completed: HaCaT keratinocytes stably expressing HRAS G12V; quantitative pERK/tERK western; 15 min primary readout, 72-h time course peaking at 6 h; antiproliferative arm in A375 (BRAF V600E) at 72 h. The discussion states "our study focused on RAS mutant cells." Evidence verdict: SUPPORTED (data generated in this paper, not imported). Bearing — direction of the finding, unchanged and load-bearing: a paradox index of 50 requires that a pERK-induction EC80 was measurable. The paper documents that encorafenib does paradoxically activate ERK and quantifies only the dose separation before it does. It is not, and the authors do not present it as, evidence of paradox absence.

[8] Adelmann — the index predicts clinical toxicity Verbatim, confirmed: "Monotherapy with vemurafenib, dabrafenib, and encorafenib induces neoplasia, most often cutaneous squamous cell carcinoma (cuSCC), at rates of approximately 22%, 6%, and 3.7%, respectively [3], [7]–[14]." Evidence: weak. Four compounds; an ecological correlation between one in vitro ratio and cross-trial clinical rates aggregated from a multi-reference block — no single denominator is recoverable. [FLAG: the prior audit quoted a concession, "averaged across multiple phase I-III trials conducted in heterogeneous patient populations." That phrase was not returned on extraction, and the paper has no formal limitations section. Verify verbatim or drop the quotation marks.] Evidence verdict: PARTIALLY-SUPPORTED. Bearing — material, unchanged: the 3.7% figure is roughly half the rate later fixed in the approved label from the randomised COLUMBUS monotherapy arm (cuSCC/KA 8%, plus new primary melanoma 5%). Any downstream citation of "3.7%" or "4%" as encorafenib's cuSCC rate is citing a pre-registrational cross-trial average that the label supersedes by about two-fold and that omits the new-primary-melanoma signal entirely.

[9] Pickles OJ et al. — DOWNGRADED Who: Pickles OJ, Drozd A, Tee L, Beggs AD, Middleton GW (University of Birmingham). Source: Paradox breaker BRAF inhibitors have comparable potency and MAPK pathway reactivation to encorafenib in BRAF mutant colorectal cancer. Oncotarget 2020;11(34):3188–3197, PMID 32922659. Venue: peer.

The claim must be split. The prior audit recorded a composite verdict of "supported"; that is claim/evidence conflation.

Component	Basis	Verdict
Time-dependent MAPK reactivation, encorafenib, BRAF-V600E lines at 3 / 24 / 48 h	Direct experiment — RealTime-Glo viability + pERK immunoblot	supported
"Group 1", "negative co-operativity", "paradoxical pathway activation in BRAF wild-type cells"	Introduction, re-cited to refs [4]–[6]	asserted-only

Verified verbatim: "Crucially, time dependent pathway reactivation was seen with PLX8394 and there was no significant difference with respect to pathway inhibition/reactivation between encorafenib and PLX8394 at any time point in any cell line." (Reported for the encorafenib arm only, per scope exclusion.) Verified cell panel — all BRAF V600E: A375, G361 (melanoma); WiDr, RKO, Colo 201, LS411N (CRC). Verified critical limitation: no BRAF wild-type line and no RAS-mutant line was tested, and the paper performed no experiment measuring paradoxical activation. CORRECTED — evidence verdict: PARTIALLY-SUPPORTED. Bearing: the paper's title places encorafenib outside the paradox-breaker category, as the comparator against it. That positioning is real and citable; the mechanistic framing in its Introduction is not this paper's evidence.

[10] Biochem J 2022 — in vivo RAF paradox in wild-type myocardium Source: Cardiomyocyte BRAF and type 1 RAF inhibitors promote cardiomyocyte and cardiac hypertrophy in mice in vivo. Biochem J. 2022;479(3):401–424, doi 10.1042/BCJ20210615, PMID 35147166. Venue: peer, primary, mouse in vivo. Verbatim: "The experimental Type 1 RAF inhibitor, SB590885, and/or encorafenib (a RAF inhibitor used clinically) increased ERK1/2 phosphorylation in cardiomyocytes, and promoted hypertrophy, consistent with a 'RAF paradox' effect, with both promoting cardiac hypertrophy in mouse hearts in vivo with increased cardiomyocyte size and no overt fibrosis." Conclusion: "Type 1 RAF inhibitors promote hypertrophy via the 'RAF paradox'." Evidence verdict: SUPPORTED. Bearing: the clearest published demonstration that encorafenib's paradox is not confined to keratinocyte tissue culture — it occurs in intact BRAF-wild-type mouse myocardium and produces a phenotype. [FLAG: two items. (a) Name the authors — the prior audit's "who" field read "Academic cardiac-signalling group", which is not a named source and fails rule 2. (b) The "and/or" construction requires figure-level confirmation of which in vivo panels used encorafenib versus SB590885.]

[11] Human-tissue case-level confirmation Who: multiple independent academic dermatology/oncology groups. Venue: peer (case reports and targeted-sequencing series). - Melanoma Res 2015, PMID 25380183 — eruptive naevi on LGX818 300 mg; "absence of the BRAF V600E mutation within a changing naevus supports the theory that BRAFi stimulates the proliferation of wild-type BRAF cells." - Melanoma Res 2016, PMID 27116335 — pyogenic granulomas at 10–16 weeks on vemurafenib or encorafenib. - Ital J Dermatol Venerol 2021, PMID 31804055 — targeted NGS of multiple second primary melanomas on encorafenib; "Our results support the paradoxical mechanism of MAPK pathway for SPMs under BRAF inhibitors." - Cureus 2024, PMID 39479087 — concurrent eruptive melanocytic naevi and multiple keratoacanthomas induced by encorafenib. - Case Rep Dermatol 2024, PMID 38831934 — eruptive melanocytic naevi on encorafenib/cetuximab/binimetinib.

Evidence: direct clinical observation with, in two instances, molecular genotyping of the induced lesion confirming BRAF wild-type status — the strongest available human-tissue evidence that the operating mechanism is paradoxical activation of the wild-type allele. Limitation: case-report tier, no denominators, no control arm. Evidence verdict: SUPPORTED (for the existence of the phenomenon; not for any rate).

5.3 — REVIEW TIER (AWARENESS ONLY — cannot establish paradox presence or absence)

[12] Ciccolini J, Milano GA — DOWNGRADED Source: Fewer cetuximab-related skin-toxicities in colorectal cancer patients treated with encorafenib: a Yin and Yang effect of ERK paradoxical activation. Ann Oncol 2022;33(9), doi 10.1016/j.annonc.2022.06.002, PMID 35718337. Venue: review / editorial commentary (confirmed). Evidence basis: cross-trial clinical observation interpreted through a mechanistic lens. No new experiment, no controlled comparison, no denominators established. CORRECTED — verdict: ASSERTED-ONLY. (Prior audit: "partially-supported". Under rule 2, review can never on its own establish that a molecule does or does not paradoxically activate. "Partially-supported" implies evidentiary weight this tier cannot carry.) Bearing — retained, because it is a paradox claim in the OPPOSITE direction: it treats encorafenib's paradoxical ERK activation in BRAF-WT skin as clinically real and consequential enough to blunt an EGFR inhibitor's on-target dermal toxicity. As signal only, it is directionally incompatible with a paradox-sparing narrative.

[13] Carr MJ, Sun J, Eroglu Z, Zager JS

"BRAF inhibition with encorafenib exhibits substantial antitumor activity with less paradoxical MAPK pathway activation leading to treatment resistance."

Source: An evaluation of encorafenib for the treatment of melanoma. Expert Opin Pharmacother 2020;21(2), PMID 31790307. Venue: review. Evidence basis: none identified. No head-to-head paradox experiment cited; "less" is unquantified; no comparator concentration or assay specified; the sentence conflates paradox with acquired resistance — two different phenomena. Verdict: ASSERTED-ONLY (unchanged). Bearing: the archetypal awareness-tier assertion a competitive narrative could mistake for evidence. This is the sentence to watch for in any competitor deck. [FLAG: quotation not independently re-retrieved at this data cut. Verify verbatim before external use.]

[14] Indini A, Mandalà M — DOWNGRADED

The encorafenib + binimetinib combination "shows peculiar pharmacodynamic properties which translate in a higher on-target potency and paradox index."

Source: Expert Opin Drug Saf 2020;19(10), PMID 32857940. Venue: review. Evidence basis: pure re-citation of the single Adelmann 2016 in vitro index. No new data. CORRECTED — verdict: ASSERTED-ONLY. (Prior audit: "partially-supported" — which double-counts one experiment by crediting both the paper that ran it and the paper that quotes it.) Bearing: framing risk. A high paradox index is a dose-separation statistic, not an absence of paradox — and here it is offered as a differentiating property of a combination whose MEK-inhibitor partner is itself the conventional pharmacological means of suppressing paradox output. [FLAG: quotation not independently re-retrieved at this data cut.]

[15] Koelblinger P, Thuerigen O, Dummer R — DOWNGRADED

Gastric hyperplasia, the mouse surrogate of cuSCC, "does not even occur at the most efficacious dose of encorafenib."

Source: Development of encorafenib for BRAF-mutated advanced melanoma. Curr Opin Oncol 2018;30(2):125–133, PMID 29356698. Venue: review. Evidence basis, verified: the gastric-hyperplasia claim is cited to reference [10] = Stuart DD et al., AACR 2012 — a conference abstract whose full text is confirmed non-retrievable (AACR HTTP 403 at this data cut) and which does not contain gastric hyperplasia in any indexed excerpt. This is a review paraphrasing an inaccessible conference abstract: two steps below grounding, with no primary source reachable at any tier. CORRECTED — independence is worse than the prior audit recorded. Verified affiliations and disclosures: Olaf Thuerigen is an employee of Pierre Fabre Pharma GmbH — the EU MAH lineage, not merely "affiliated with the sponsor side". Reinhard Dummer, the COLUMBUS lead investigator, discloses consulting/advisory relationships with Novartis and Pierre Fabre among others. CORRECTED provenance of the cuSCC figure: Koelblinger quotes 4% for encorafenib, cited to reference [7] = Delord 2017 (the phase I dose-escalation paper) — not to Adelmann. The prior audit treated "3.7% (Adelmann)" and "4% (Koelblinger)" as interchangeable; they have different provenance and different denominators. Both remain roughly half the label figure. CORRECTED — verdict: ASSERTED-ONLY. (Prior audit: "partially-supported".) Note: Koelblinger also describes encorafenib within the "so called αC-OUT BRAFi" class that "promote RAF–RAS–GTP interaction and RAF dimerization" (citing Karoulia 2017, Nat Rev Cancer — itself a review). Even the sponsor-adjacent review literature does not position encorafenib as paradox-exempt.

5.4 — CONFERENCE TIER

[16] Stuart DD et al., AACR 2012 abstract 3790 — SCOPE REDUCED Source: Abstract 3790: Preclinical profile of LGX818: A potent and selective RAF kinase inhibitor. Cancer Res 2012;72(8_Suppl):3790, AACR 103rd Annual Meeting. Abstract number 3790 CONFIRMED. Full text confirmed non-retrievable — AACR HTTP 403 at this data cut.

CORRECTED — what this abstract actually supports:

Attributed fact	Status
"inactive against BRAF wild-type tumors at doses up to 300 mg/kg bid"	verified in indexed abstract text
Tumour regression in BRAF-mutant xenografts at doses as low as 1 mg/kg	verified in indexed abstract text
">400 cell lines expressing wild-type BRAF" screen	NOT in retrievable text — secondary attribution only
>30 h dissociation half-life	NOT in retrievable text — secondary attribution only
Mouse gastric hyperplasia surrogate	NOT in retrievable text — secondary attribution only
pERK EC50 3 nM / antiproliferative EC50 4 nM in A375	NOT in retrievable text — secondary attribution only

The prior audit sourced all six to this abstract. Four reach the file only through review paraphrase of a document nobody in this audit could open.

Verdict: ASSERTED-ONLY. Bearing — THE CENTRAL CLAIM/EVIDENCE DIVERGENCE IN THE FILE, and it survives intact: the verified content is an antiproliferative and xenograft growth screen. A proliferation readout is structurally incapable of detecting paradoxical pERK induction. Paradox in a BRAF-WT cell manifests as pathway activation and, over time, neoplasia in RAS-lesioned keratinocytes — not as growth inhibition of an unlesioned wild-type line. When this dataset is used to argue that encorafenib does not paradoxically activate, the evidence does not reach the claim.

5.5 — COMPANY TIER (AWARENESS ONLY)

[17] Array BioPharma press release, 1 June 2015 — SPLIT Source: "Binimetinib And Encorafenib Combination Shows Promising Clinical Activity And Potential Differentiated Safety In BRAF-Mutant Melanoma", PR Newswire 300091510. Venue: company. - Phase 1b/2 dose-escalation figures (N=55 BRAF-naive; pyrexia and photosensitivity 11% each at 400/450 mg; ORR 75%, 41/55; mPFS 11.3 months): company-tier reporting of company-run data — awareness only. - "wide therapeutic index" nonclinical assertion: ASSERTED-ONLY, traced to the non-retrievable Stuart abstract.

CORRECTED — verdict: ASSERTED-ONLY / awareness. (Prior audit: "partially-supported".) NEGATIVE FINDING OF RECORD (the genuinely valuable content, recorded separately from any verdict): this release makes no claim about paradoxical activation, cuSCC, keratoacanthoma, RAF dimers or paradox breaking. The sponsor's differentiation pitch was tolerability — pyrexia and photosensitivity — not paradox.

[18] Pfizer BRAFTOVI HCP promotional site — DOWNGRADED Source: braftovi.pfizerpro.com/mektovi-m/mechanism-of-action; /c/mechanism-of-action; pfizermedical.com/braftovi/clinical-pharmacology. Venue: company. CORRECTED — verdict: AWARENESS-ONLY. (Prior audit: "supported" — a company-tier source cannot carry an evidentiary verdict under rule 2, in either direction, even when the observation is sound.) Evidence basis: the "long binding time" claim is referenced only to the prescribing information, not to a mechanism study; and the underlying >30 h number appears at label tier in the EU SmPC only — the FDA label omits it entirely (re-verified). NEGATIVE FINDING OF RECORD: the sponsor's own promotional MOA pages carry the paradoxical-activation statement and make no "paradox breaker", "dimer-selective", "equipotent monomer/dimer" or "no RAF transactivation" claim, and no comparative paradox differentiation. The company venue is not the source of a paradox-sparing narrative for this molecule.

5.6 — SYSTEMATIC NEGATIVE

[19] No source in any venue applies "paradox breaker", "dimer-selective", "equipotent against monomer and dimer", "no RAF transactivation", or "does not induce paradoxical activation" to encorafenib.

CORRECTED — venue: AUDIT / SEARCH RECORD. (Prior audit tagged this "peer". It is not. Peer papers are cited inside it, but a negative search result cannot inherit peer tier from its constituents. This is the venue-integrity error the brief identifies as the most damaging class.)

Search executed: FDA USPI rev. 2/2026 (retrieved, read); EU SmPC (retrieved, read); PubMed encorafenib AND paradoxical (21 records at 1 Sep 2026, per the prior audit — [FLAG: this query count was not independently re-executed here]); AACR 2012 abstract 3790 (403); Array 2015 PR; Pfizer HCP MOA pages; Pierre Fabre and Pfizer corporate materials.

Positioning stated affirmatively in peer literature: Pickles 2020 (title contrasts paradox breakers with encorafenib as the comparator); Cotto-Rios 2020 (encorafenib used as the tool compound because it half-occupies dimers).

CORRECTED — evidence verdict: SUPPORTED as a negative search result. (Prior audit: "contradicted" — again the narrative axis in the evidence field.) Bearing: the contradicting primary evidence is affirmative, not merely absent — Cotto-Rios 2020 (class II, class III, NRAS), Adelmann 2016 (finite pERK-induction EC80 in HRAS-G12V keratinocytes), Biochem J 2022 (in vivo ERK1/2 induction, wild-type mouse myocardium), and encorafenib's own FDA label §5.2. [FLAG: a negative result cannot be proven exhaustive. Paywalled congress archives (AACR/ASCO/ESMO, 403) and internal investor decks were not accessible. State this limitation wherever the negative is cited.]

6. Clinical paradox signal

STRONGLY CONSISTENT WITH PARADOXICAL ACTIVATION. Encorafenib produces the canonical paradox toxicity spectrum in humans; the clinical record contains no signal of paradox absence.

Label-tier rates — FDA USPI §5.1, rev. 2/2026, re-retrieved verbatim

Regimen	cuSCC / KA	BCC	New primary melanoma	Other
COLUMBUS, encorafenib + binimetinib	2.6%	1.6%	—	median time to first cuSCC/KA 5.8 months (range 1–9)
COLUMBUS, encorafenib SINGLE AGENT	8%	1%	5%	—
BEACON CRC, + cetuximab	1.4%	—	1.4%	—
PHAROS NSCLC, + binimetinib	2% (cuSCC)	—	—	skin papilloma 2%
BREAKWATER, + cetuximab + mFOLFOX6	SCC 0.9%; KA 0.4%	1.3%	melanoma in situ 0.4%	melanocytic naevus 5.6%; skin papilloma 3%
BREAKWATER, + cetuximab + FOLFIRI	KA 1.4%	—	—	skin papilloma 2.8%

The single-agent row is the paradox-relevant one. The combination figures are confounded by the MEK inhibitor, which is the conventional pharmacological means of suppressing paradox output.

[FLAG: denominators. FDA §5.1 states no n for any of these rows. The prior audit presented "n=192" for the single-agent arm as though label-tier; 192 comes from Gogas 2019 (peer), not from §5.1. A label-extraction pass returned candidate denominators (single-agent 194, BEACON 216, PHAROS 98, BREAKWATER 232 / 71) as a derived table, not verbatim §5.1 text. Every denominator in this section must be re-confirmed against USPI §6.1 / §14 before external use.]

Peer-tier safety population

Gogas HJ, Flaherty KT, Dummer R, et al. Adverse events associated with encorafenib plus binimetinib in the COLUMBUS study. Eur J Cancer 2019;119:97–106, PMID 31437754 — n=570 total (COMBO450 192 / ENCO300 192 / VEM 186); hyperkeratosis with COMBO450 grade 1/2/3 = 17%/5%/1%. Exposure durations differ materially by arm (51 vs 31 vs 27 weeks), which biases naive cross-arm cuSCC comparison in encorafenib's favour on the monotherapy arm. [FLAG: Gogas full text returned HTTP 410 at this data cut; these figures are carried forward from the prior audit unverified.]

CORRECTED — an "unknown" that is a silence, not an absence

The prior audit recorded the COLUMBUS vemurafenib-arm cuSCC/KA rate as UNKNOWN, on the basis that the Gogas abstract does not report it by arm and the full text is paywalled.

That unknown was not verified to be absent — it was not reached. At this data cut, indexed abstract text for the COLUMBUS Lancet Oncology report surfaces a vemurafenib-arm figure in the region of 26% for cutaneous squamous-cell carcinomas, the majority keratoacanthoma type. [FLAG: I do NOT adopt this number. Lancet Oncology (Dummer 2018, PMID 29573941) and the JCO COLUMBUS 5-year update (10.1200/JCO.21.02659) both returned HTTP 403/cookie walls to me. A human with journal access must retrieve the arm-by-arm cutaneous-malignancy table. Recorded as NOT RETRIEVED HERE, not as unknown — rule 4 requires an honest unknown to be a verified absence.] This matters: the randomised within-trial encorafenib-versus-vemurafenib cuSCC contrast is the cleanest clinical paradox comparison available, and it is probably obtainable.

Material discrepancy to carry forward

The figure most often quoted for encorafenib's cuSCC rate in reviews and comparative narratives is 3.7% (Adelmann 2016, multi-reference block) or 4% (Koelblinger 2018, cited to Delord 2017 phase I). The approved label's randomised single-agent figure is 8% cuSCC/KA plus a separate 5% new primary melanoma. The circulating literature figure understates the label figure by roughly two-fold and omits the new-primary-melanoma signal entirely. Any competitive comparison built on 3.7–4% is built on a pre-registrational cross-trial average without denominators.

Regulatory behavioural requirements — the clearest institutional statement that paradox is operative

FDA §5.1, verbatim: "Perform dermatologic evaluations prior to initiating treatment, every 2 months during treatment, and for up to 6 months following discontinuation of treatment." [CORRECTED: the prior audit attributed this schedule to EU §4.4 as well. It is NOT in the retrieved EU §4.4 text. Do not cite EU §4.4 for the dermatologic schedule.]
EU §4.4, verbatim: "Patients receiving encorafenib should undergo a head and neck examination, chest/abdomen computerised tomography (CT) scan, anal and pelvic examinations (for women) and complete blood cell counts prior to initiation, during and at the end of treatment as clinically appropriate."
FDA: "Discontinue BRAFTOVI for RAS mutation-positive noncutaneous malignancies." EU: "It should be considered to permanently discontinue encorafenib in patients who develop RAS mutation-positive non-cutaneous malignancies."

RAS-mutant tumour progression on drug — UNKNOWN (verified as absent)

No published case series or trial-level rate of RAS-mutant malignancy progression on encorafenib was identified. Both labels mandate surveillance and discontinuation without disclosing any observed incidence. This is a genuine, verified absence.

Inverse-direction clinical signal

Reduced cetuximab-related skin toxicity in CRC patients on encorafenib, attributed by Ciccolini & Milano to paradoxical ERK activation in BRAF-wild-type keratinocytes counteracting EGFR-blockade apoptosis. Awareness tier only — but directionally incompatible with a paradox-sparing profile.

Bottom line

The clinical readouts are consistent with paradoxical activation, at a lower absolute rate than vemurafenib. Lower rate, same mechanism. Nothing in the clinical record supports a categorical difference in kind — and the quantitative comparison that would test "lower rate" most cleanly (randomised within-COLUMBUS, encorafenib versus vemurafenib) has not been retrieved here.

7. Unknowns and not-disclosed

No encorafenib–BRAF co-crystal structure exists. Marini 2022 states it verbatim; RCSB full-text search returns exactly one entry (9BP8, BRAF + PF-07799933, 1.73 Å), in which encorafenib is not the ligand. Every binding-mode statement for encorafenib is a docked pose on the BRAF–dabrafenib template PDB 5CSW. No structural evidence exists of how encorafenib occupies a RAF dimer. (Verified absence.)
Quantitative monomer-versus-dimer potency is NOT DISCLOSED. No side-by-side IC50 against monomeric versus dimeric BRAF V600E, no dimer-selectivity ratio, no measured cooperativity coefficient for encorafenib's own second-site binding. The label's 0.35 / 0.47 / 0.30 nM values are cell-free enzyme IC50s and carry no monomer/dimer information. Clayton 2025 (eLife 13:RP95334) names encorafenib once, Introduction only, with no values; Tkacik 2026 (eLife RP110344) did not test it. (Verified absence across both candidate sources.)
Activity against BRAF FUSION proteins is UNKNOWN. No encorafenib data in any dimer-constitutive fusion model (KIAA1549-BRAF or equivalent) identified in any venue. (Verified absence.)
No primary experiment links residence time to PARADOX behaviour. The EU label attaches the >30 h figure only to prolonged pERK inhibition in target cells — and Koelblinger specifies it is measured from V600E-mutant BRAF, i.e. on the mutant target, not on wild-type BRAF or CRAF where paradox occurs. No study varies residence time and measures paradoxical pERK induction as the dependent variable. The residence-time-to-lower-toxicity link exists only as a hedged review-tier hypothesis.
Stuart AACR 2012 abstract 3790 full text is non-retrievable (HTTP 403). Four of the six facts the prior audit sourced to it — the >400-cell-line screen, the >30 h off-rate, the gastric-hyperplasia surrogate, and the A375 pERK/antiproliferative EC50s — are not in the retrievable text and reach the file only through review paraphrase. Comparator off-rates (2 h dabrafenib, 0.5 h vemurafenib) share this dependency.
No trial-level or series-level rate of RAS-mutant tumour progression on encorafenib is published, despite both labels mandating surveillance and discontinuation. The regulators require the behaviour without disclosing an incidence. (Verified absence.)
No paradox-breaker / dimer-selective / equipotent-monomer-and-dimer / no-RAF-transactivation claim for encorafenib exists in any venue. Systematic negative, not an unexamined gap — but not provably exhaustive: paywalled congress archives (AACR/ASCO/ESMO, 403) and internal investor decks were not accessible.
The EU label's silence on paradox is unexplained. Whether it reflects substantive regulatory disagreement, differing labelling conventions, or a drafting artefact is UNKNOWN from public documents. No EMA assessment-report rationale for the omission was located. [FLAG: the prior audit's per-term hit counts across the full 63-page EPAR Annexes I–IIIB were NOT reproducible — the EMA PDF would not text-extract at this data cut. My zero-hit verification for "paradox"/"paradoxical"/"dimer" is on the UK emc SmPC (Annex I equivalent) only.]
Whether encorafenib's paradox differs QUANTITATIVELY from dabrafenib in a controlled setting is only weakly established — a single four-compound in vitro ratio (paradox index 50 vs 10) plus cross-trial cuSCC rates without denominators. No randomised head-to-head encorafenib-versus-dabrafenib trial reporting cuSCC as an endpoint was identified.
COLUMBUS vemurafenib-arm cuSCC/KA — NOT RETRIEVED HERE (downgraded from "unknown"). A vemurafenib-arm figure appears to exist in the published record; Lancet Oncology and the JCO 5-year update were both inaccessible to me (403). [FLAG: retrieve with journal access. This is the cleanest available randomised clinical paradox contrast and it is probably obtainable.]
Plixorafenib / FORE8394 / PLX8394 excluded by instruction. Where source papers used it as the comparator arm (Pickles 2020; Adelmann 2016), only encorafenib-arm data were extracted. No comparison against it is offered or implied.

8. Citations, by tier

LABEL - BRAFTOVI (encorafenib) US Prescribing Information, rev. 2/2026 — §1, §5.1, §5.2, §12.1. labeling.pfizer.com id=12990; cf. FDA accessdata 210496s017lbl.pdf. [Retrieved and read 1 Sep 2026.] - Braftovi (encorafenib) 75 mg hard capsules, Summary of Product Characteristics — §4.4, §5.1; ATC L01EC03; MAH Pierre Fabre Limited. medicines.org.uk/emc/product/9500/smpc. [Retrieved and read 1 Sep 2026.] - Braftovi EPAR Product Information (Annexes I–IIIB), EMA. [FLAG: NOT text-extractable at this data cut — binary FlateDecode. Any claim sourced to the EPAR specifically, including the MAH entity and the per-term hit counts, is unverified.]

PEER — primary experiment - Cotto-Rios XM, Agianian B, Gitego N, Zacharioudakis E, Giricz O, Wu Y, Zou Y, Verma A, Poulikakos PI, Gavathiotis E. Inhibitors of BRAF dimers using an allosteric site. Nat Commun 2020;11:4370. doi 10.1038/s41467-020-18123-2. PMC7462985. - Adelmann CH, Ching G, Du L, Saporito RC, Bansal V, Pence LJ, Liang R, Lee W, Tsai KY. Comparative profiles of BRAF inhibitors: the paradox index as a predictor of clinical toxicity. Oncotarget 2016;7(21):30453–30460. doi 10.18632/oncotarget.8351. PMID 27028853. PMC5058692. - Pickles OJ, Drozd A, Tee L, Beggs AD, Middleton GW. Oncotarget 2020;11(34):3188–3197. doi 10.18632/oncotarget.27681. PMID 32922659. PMC7456617. - Cardiomyocyte BRAF and type 1 RAF inhibitors promote cardiomyocyte and cardiac hypertrophy in mice in vivo. Biochem J 2022;479(3):401–424. doi 10.1042/BCJ20210615. PMID 35147166. [FLAG: name the authors.] - Marini E, Marino M, Gionfriddo G, Maione F, Pandini et al. Investigation into the Use of Encorafenib to Develop Potential PROTACs Directed against BRAF V600E Protein. Molecules 2022;27(23):8513. doi 10.3390/molecules27238513. PMID 36500607. PMC9736157. - Tkacik E, Jang DM, Boxer K, Ha BH, Eck MJ. eLife Reviewed Preprint RP110344, v1, 23 March 2026. doi 10.7554/eLife.110344.1 — encorafenib NOT tested; cited here only to refute a class-exclusivity generalisation. - Clayton J, Romany A, Matenoglou E, Gavathiotis E, Poulikakos PI, Shen J. Mechanism of dimer selectivity and binding cooperativity of BRAF inhibitors. eLife 2025;13:RP95334. PMID 39945510. (Corrected from bioRxiv preprint citation.) Contains no encorafenib IC50, cooperativity or selectivity value. - Allosteric coupling asymmetry mediates paradoxical activation of BRAF by type II inhibitors. eLife 95481.

PEER — clinical - Delord JP, et al. Clin Cancer Res 2017;23(18):5339–5348. doi 10.1158/1078-0432.CCR-16-2923. PMID 28611198. - Gogas HJ, Flaherty KT, Dummer R, et al. Eur J Cancer 2019;119:97–106. PMID 31437754. [FLAG: HTTP 410 at this data cut.] - Dummer R, Ascierto PA, Gogas HJ, et al. Lancet Oncol 2018;19(5):603–615. PMID 29573941. [FLAG: not retrieved — 403/cookie wall.] - COLUMBUS 5-Year Update. J Clin Oncol. doi 10.1200/JCO.21.02659. [FLAG: not retrieved — 403.]

PEER — case level - Melanoma Res 2015, PMID 25380183 · Melanoma Res 2016, PMID 27116335 · Ital J Dermatol Venerol 2021, PMID 31804055 · Cureus 2024;16(9):e70540, PMID 39479087 · Case Rep Dermatol 2024, PMID 38831934.

CONFERENCE - Stuart DD, Li N, Poon DJ, et al. Abstract 3790: Preclinical profile of LGX818. Cancer Res 2012;72(8_Suppl):3790. [Full text non-retrievable, HTTP 403, confirmed at this data cut.]

REVIEW — awareness only, cannot establish paradox - Ciccolini J, Milano GA. Ann Oncol 2022;33(9). PMID 35718337. - Koelblinger P, Thuerigen O, Dummer R. Curr Opin Oncol 2018;30(2):125–133. PMID 29356698. (Thuerigen: employee, Pierre Fabre Pharma GmbH.) - Carr MJ, Sun J, Eroglu Z, Zager JS. Expert Opin Pharmacother 2020;21(2). PMID 31790307. - Indini A, Mandalà M. Expert Opin Drug Saf 2020;19(10). PMID 32857940. - Karoulia Z, Gavathiotis E, Poulikakos PI. Nat Rev Cancer 2017;17:676–691. - Roskoski R Jr. Pharmacol Res 2018;135:239–258. PMID 30118796.

COMPANY — awareness only - Array BioPharma press release, 1 June 2015, PR Newswire 300091510. - Array BioPharma, "Array Announces Agreement To Acquire ENCORAFENIB (LGX818)", investor.arraybiopharma.com — $85M upfront from Novartis Pharma AG. - Pfizer BRAFTOVI HCP MOA pages; pfizermedical.com/braftovi/clinical-pharmacology.

REGISTRY / STRUCTURAL - ChEMBL CHEMBL3301612 — synonyms and research codes; first approval 2018. - RCSB PDB search API, full-text "encorafenib", 1 Sep 2026 — total_count 1, entry 9BP8 (BRAF + PF-07799933, 1.73 Å); encorafenib not present as ligand.

AUDIT / SEARCH RECORD (not peer — corrected) - PubMed query encorafenib AND paradoxical, 21 records at 1 Sep 2026. [FLAG: count carried forward from the prior audit, not independently re-executed here.]

9. What changed, and why it matters

#	Attack vector	Finding
1	Venue integrity	Systematic-negative claim was tagged peer; it is a search record. A V600E/D/K statement was attributed to the FDA label; it is the EU SmPC. Two company/review entries carried evidentiary verdicts they cannot support.
2	Claim/evidence conflation	Pickles downgraded — the paper tested no wild-type and no RAS-mutant line and ran no paradox experiment. Three review-tier entries downgraded from partially-supported to asserted-only. Stuart's abstract shrank from six attributed facts to two.
3	Identity	ARRY-085 / ARRY-424704 struck — unverifiable. EU MAH entity flagged.
4	Absence vs silence	EU label actually retrieved and read — paradox absence confirmed on Annex I, per-term counts flagged as unreproducible. RCSB actually searched — no co-crystal confirmed, but the search is not null. The COLUMBUS vemurafenib-arm rate reclassified from "unknown" to "not retrieved here".
5	Overreach	The αC-OUT class-exclusivity argument refuted by Tkacik 2026 (type I and type II both paradox-activate). The structural argument should be retired, not caveated.

And the correction that cuts the other way: the prior audit under-read its own strongest evidence. Both labels state, at label tier, that encorafenib inhibits wild-type BRAF at 0.47 nM and CRAF at 0.30 nM against 0.35 nM for BRAF V600E — no selectivity for the mutant target, and CRAF inhibited more potently than it. That single fact needs no structure, no docking, no review and no inference. It is the biochemical substrate for paradox, it is stated by both regulators, and it is the sentence to lead with.

mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026

Verification confidence: high

Mosperafenib — Paradox-Claim Audit (adversarially re-verified)

Data cut: 1 September 2026. Plixorafenib (FORE8394 / PLX8394) is out of scope and was not researched, inferred about, or compared against.

Venue tiers. label and peer are grounding. conference is provisional. company and review are awareness only and can never on their own establish that a molecule does or does not cause paradoxical activation.

Status of this document. This is a refutation pass over a prior audit. Every label and peer tag was re-checked against source; three sections were rewritten after the underlying source contradicted the prior reading. Corrections carrying operational consequence are marked [CORRECTED]; residual human-verification points are marked [FLAG].

1. Resolved identity

RESOLVED. INN: mosperafenib. Code numbers: RG6344, RO7276389. Vendor alias "B-Raf IN 2". CAS 2649372-20-1. C20H17F2N5O4S, MW 461.44. (3R)-N-{2-cyano-4-fluoro-3-[(3-methyl-4-oxo-3,4-dihydroquinazolin-6-yl)oxy]phenyl}-3-fluoropyrrolidine-1-sulfonamide — a quinazolinone aryl-ether / pyrrolidine-sulfonamide.

Originator and current sponsor: F. Hoffmann-La Roche Ltd. No out-licensing or partnering identified.

[CORRECTED] The identity bridge is stronger than previously recorded.

peer — Mol Cancer Ther 2026;25(4):599-609 states verbatim in its abstract: "the next-generation BRAFi mosperafenib (RG6344/RO7276389)". The mosperafenib = RG6344 = RO7276389 equivalence is therefore peer-reviewed, not conference-only. (Also at conference tier in ASCO 3542 and AACR CT017.)
conference — the link to "Compound Ia" rests on AACR 2025 Abstract 5606, verbatim: "a next generation brain permeable, paradox breaker BRAFi (RG6344, previously identified as Compound Ia in Wichmann et al., Clin Cancer Res 2022)". This single sentence remains the only explicit Compound Ia ↔ RG6344 statement located.
peer — but the bridge is partly corroborated at peer tier: Cancer Res Commun 2026 (open access, full text read) states "Mosperafenib is a novel, brain-penetrant BRAFi designed to overcome this paradox, showing promising activity in preclinical models (6)", and its reference 6 is Wichmann, Clin Cancer Res 2022 (verified in the reference list of the retrieved JATS full text). A peer-reviewed paper thus attributes mosperafenib's preclinical characterization to the Wichmann paper, even though it does not use the string "Compound Ia".

[FLAG] The prior audit asserted that Wichmann's "Compound Ia" and Bonfill-Teixidor's "compound 1a (C1a)" (Cancer Res 2022;82:2552-2564) are the same entity. No located source states this. Shared authorship (Wichmann, Handl, Bischoff, Pettazzoni) makes it near-certain, but it is an inference and a human must confirm it before the melanoma-brain-metastasis dataset is attributed to this molecule.

Regulatory status — LABEL TIER IS EMPTY (verified by retrieval, not by assumption)

Query	Endpoint	Result (re-run 1 Sep 2026)
`openfda.generic_name:"mosperafenib"`	api.fda.gov/drug/label.json	`NOT_FOUND`
free-text `mosperafenib`	api.fda.gov/drug/label.json	`NOT_FOUND`
free-text `RG6344`	api.fda.gov/drug/label.json	`NOT_FOUND`
`mosperafenib`	ClinicalTrials.gov API v2	0 studies
`RO7276389`	ClinicalTrials.gov API v2	0 studies
`RG6344`	ClinicalTrials.gov API v2	0 studies

No approved label exists anywhere. No label-grounded paradox claim can exist for this molecule.

[FLAG] [CORRECTED] EMA absence is not established. The prior audit reported a 401 and treated the matter as searched. The EMA medicine finder is JS-driven and did not return a conclusive result here either. Record this as not retrieved, not as searched and absent. Given no approval anywhere and a phase I asset now dropped (§7), an SmPC is implausible — but implausible is not verified.

Clinical development

Single study: ISRCTN13713551 (verified via the ISRCTN WHO-format API). Phase IA/B, open-label, multicentre, dose-escalation + cohort expansion of RO7276389 alone and with cobimetinib. Primary sponsor "Roche (United States)"; support F. Hoffmann-La Roche. Registered 03/12/2021; target 292; recruitment status "No longer recruiting". Eligibility is BRAF V600 mutation-positive advanced solid tumours, or BRAF V600-mutant melanoma with CNS metastases — no RAS-mutant population was ever eligible. Not on ClinicalTrials.gov, so ctgov-only screening misses this asset entirely.

2. Binding mode — NOT DISCLOSED

No co-crystal, no PDB ID, no type I / type I½ / type II assignment, no DFG or αC-helix or R506 status, no key contact residues, no mutational SAR mapping paradox-breaking behaviour to specific residues.

review — The Acta Pharm Sin B 2024 review (PMC11120325), which tabulates DFG/αC/R506 configuration for every RAF inhibitor it covers, enters "NA" in the conformation column for Compound Ia while populating it for the comparator paradox breakers (PLX8394 and PLX7904: "DFG-in / αC-helix-out / R506-out"; hybrid 6: "DFG-out / αC-helix-in / R506-out").

[CORRECTED — scope of that inference. A review's blank field shows only that those authors did not populate it. It does not establish that no assignment exists anywhere. The prior audit called it "independent confirmation that the information is not public"; the defensible statement is "a specialist review actively tabulating this field left it empty."]

Chemotype caution retained and endorsed. The sulfonamide scaffold superficially resembles the PLX-series paradox breakers. There is no public evidence for an αC-out / R506-out mode and it must not be assumed.

[CORRECTED] The patent-derived assay is NOT evidence about this molecule — see §5, claim 9. WO2021116050A1 expressly disclaims mosperafenib. Its HTRF pERK A375/HCT116 design is a legitimate paradox assay, but it belongs to a series from which this compound is carved out.

3. Monomer versus dimer — NOT DISCLOSED, with two corrections

No monomer-versus-dimer potency comparison, selectivity ratio, forced-dimer comparison, BRAF-fusion or p61-splice-variant potency, or dimer-disruption assay (co-IP, NanoBiT/split-luciferase, BRET, crosslinking) has been published in readable form for this molecule. The phrases "dimer-selective", "equipotent against monomer and dimer" and "no RAF transactivation" appear nowhere in quantified form.

[CORRECTED] "No dimerisation assay has been reported" is too flat. The Acta Pharm Sin B dimer column is explicitly headed "Effects on RAF dimers (confirmed experimentally)" and enters "BRAF/BRAF" for Compound Ia, citing refs 146 and 147 — verified as Bonfill-Teixidor 2022 and Wichmann 2022. Correct formulation: no dimer assay is publicly readable; an independent review asserts an experimentally-confirmed BRAF/BRAF dimer effect and points to the two paywalled 2022 papers as its source. review tier: awareness only, and a pointer for paid retrieval.

The same review records no BRAF/CRAF or CRAF/CRAF entry for Compound Ia, whereas it does for PLX8394 (BRAF/CRAF, BRAF/BRAF) and BDTX4933 (all three). Whether that blank means "tested and negative" or "never tested" cannot be determined.

[CORRECTED] The affinity figures ARE traceable, and the review corroborates the pattern.

The Acta Pharm Sin B column headers, decoded from the retrieved full text, are: Conformation binding to RAF | Cell-free assay (IC50: nmol/L): ARAF, BRAF, CRAF, BRAF V600E | Effects on RAF dimers (confirmed experimentally) | Preclinical effects | Phase | Ref. The Compound Ia row therefore reads:

ARAF	BRAF (WT)	CRAF	BRAF V600E	Dimer target	Phase
NA	<5E-10	NA	1.77E-09	BRAF/BRAF	Preclinical

company (vendor) — ProbeChem, MedKoo, Selleck, MedChemExpress list Kd 0.6 nM BRAF WT / 1.2 nM BRAF V600E / ~1.7 nM CRAF, plus (previously unrecorded) kinase IC50 <1.77 nM across BRAF mutants (V600E/K/A/D) and cytotoxic activity 5.2-30.2 nM across 94 cell lines. The ProbeChem datasheet cites a primary source: Wichmann J, et al. Clin Cancer Res. 2022;28(4):770-780. The prior audit's statement that these figures "carry no primary citation" and "could not be traced" is wrong.

Analytic flag, now better supported. Two independent secondary sources — a vendor datasheet and an academic review — both describe a compound that engages wild-type BRAF at least as tightly as BRAF V600E, at sub-to-low-nanomolar affinity. That is a pan-RAF-like binding profile. It is not by itself disqualifying (a PLX-class paradox breaker breaks the paradox by disrupting the dimer interface after binding, not by failing to bind wild-type RAF), but it means the paradox claim rests entirely on dimer-disruption behaviour that has never been directly measured for this molecule in any readable public source. Both attestations trace to a paywalled paper; neither is grounding.

4. Corrected reading of the primary paradox data (MCT 2026 supplement)

This is the only genuinely mosperafenib-specific, publicly readable paradoxical-activation dataset in existence, so the read has to be right.

Suppl Fig S2 — pERK1/2 in three BRAF wild-type CRC lines

Densitometry of Fig 3A/B/C. SW48 (BRAF WT / KRAS WT), HCT116 (BRAF WT / KRAS mut), SW480 (BRAF WT / KRAS mut), at 10 / 100 / 1000 nM, each panel run twice: monotherapy and + cetuximab 5 µg/mL, normalised to the overall DMSO control.

[CORRECTED] The prior audit read only the monotherapy half of every panel. Full read (mosperafenib / encorafenib, fold vs DMSO control):

Line	Block	DMSO	10 nM	100 nM	1000 nM
SW48	mono	1.0 / 0.85	1.15 / 1.35	1.45 / 2.25	1.6 / 3.65
SW48	+cetux	0.45 / 0.9	0.6 / 0.8	0.85 / 0.95	0.8 / 1.0
HCT116	mono	1.1 / 1.1	2.7 / 3.35	2.85 / 4.45	0.95 / 2.05
HCT116	+cetux	2.2 / 2.15	3.0 / 3.2	3.65 / 4.65	1.85 / 2.6
SW480	mono	1.0 / 1.1	1.45 / 2.7	2.05 / 2.85	1.15 / 0.95
SW480	+cetux	1.75 / 2.2	1.85 / 3.0	1.95 / 3.65	1.95 / 1.85

Three consequences:

The peak mosperafenib pERK induction is ~3.65× DMSO (HCT116, 100 nM, + cetuximab) — not ~2.85×. Relative to the cetuximab-matched DMSO control (~2.2×) the same point is ~1.7×; both framings are given because the figure's own normalisation is to the overall control.
Mosperafenib is not uniformly below encorafenib. In SW480 at 1000 nM it is the higher bar in both blocks (mono 1.15 vs 0.95; +cetux 1.95 vs 1.85).
The cetuximab block is the clinically relevant CRC regimen and its omission understated the induction in exactly the condition the programme is being developed in.

Retained limitations: n and replicate structure not stated; densitometry of immunoblots is semi-quantitative; three CRC lines only; no NRAS- or HRAS-mutant line; no melanocyte, keratinocyte or skin model — the tissue where the class toxicity actually occurs; no statistical testing; y-axis labelled "% vs control" while plotting fold-change; and the caption frames the experiment as assessing "the paradoxical induction of the MAPK pathway by encorafenib**", i.e. as an encorafenib characterisation.

[CORRECTED] Suppl Fig S3 — missed, and it bears directly on the question

The prior audit stated "no other figure addresses paradoxical activation." Suppl Fig S3 does. HT29 (BRAF V600E), mosperafenib or encorafenib at 300 nM, ± cetuximab 5 µg/mL, ± EGF 10 ng/mL for 15 min, time course 0 / 3 / 24 / 48 h, immunoblotted for P-ERK, ERK, P-c-RAF, c-RAF, P-AKT (Vinculin / GAPDH loading).

Reading the blot: under mosperafenib monotherapy, P-ERK is suppressed at 3 h but rebounds markedly at 24 h and strongly at 48 h; and P-c-RAF is induced at 24-48 h, in both the −EGF and +EGF arms. Encorafenib behaves similarly. Cetuximab co-treatment substantially blunts both.

[FLAG] This is my own read of band intensity. No densitometry is provided for S3, it is a single blot, and n is not stated. A human must verify against Fig 3 and the full text before this is used externally.

Interpretive discipline: P-ERK rebound in a V600E line is the well-known EGFR-mediated feedback reactivation that motivates BRAFi + anti-EGFR combinations — it is not classical paradoxical dimer activation in wild-type tissue, and must not be conflated with it. What is nonetheless material is that P-c-RAF is induced under mosperafenib — an on-pathway RAF-activation readout for this molecule that the prior audit recorded as non-existent.

Suppl Fig S1 — V600E potency (corrected and completed)

peer — HTRF pERK, BRAF V600E lines only. No wild-type or RAS-mutant arm exists in this figure.

	Mosperafenib	Encorafenib
LS411N IC50 / IC80 / IC95 (nM)	10.8 / 31.87 / 100.1	8.13 / 29.18 / 105.5
HT29 IC50 / IC80 / IC95 (nM)	9.79 / 25.14 / 72.57	6.3 / 19.37 / 68.43

[CORRECTED] Encorafenib is marginally more potent at IC50 and IC80 in both lines, but not at IC95 in LS411N, where mosperafenib is marginally more potent (100.1 vs 105.5). Note also an internal inconsistency: the caption reports "IC50, IC90, and IC95" while the table gives IC80.

Full supplementary set enumerated via the figshare API (collection 8398399): S1-S9 + Supplementary Table 1 (10 items). S4-S8 are in vivo tumour volumes, body weights, bevacizumab combination, PDX clinicopathology and Kaplan-Meier curves; S9 is mosperafenib + binimetinib + FOLFOX in RKO (BRAF V600E); Table 1 is PDX whole-exome variant calls. S1, S2 and S3 are the mechanism-relevant items.

5. Paradox claims

Claim 1 — "does not trigger pERK overactivation in BRAF wild-type contexts"

Venue: peer — Mol Cancer Ther 2026;25(4):599-609. DOI 10.1158/1535-7163.MCT-25-0562. PMID 41340484. Online 3 Dec 2025.
Who: Renner F, Eckmann J, Handl C, Wichmann J, Schnetzler G, Kratochwil NA, Keshelava N + VHIO (Ros J, Palmer HG, Pettazzoni P et al.). Roche-sponsored.
Evidence: Fig 3A-C / Suppl Fig S2, read directly (§4). Mosperafenib raises pERK above baseline in all three BRAF wild-type lines, peaking at ~3.65× DMSO (HCT116, 100 nM + cetuximab) and ~2.85× in monotherapy, with the classic bell-shaped concentration-response that is the signature of paradoxical activation, resolving only at 1000 nM in monotherapy. Consistently less than encorafenib at most matched points, but not absent.
Verdict: CONTRADICTED (unchanged; the contradiction is now larger than previously recorded).

Claim 2 — comparative form: substantially LESS paradoxical activation than encorafenib

Venue: peer — same source, Fig 3A-C / Suppl Fig S2.
Evidence: Direction holds at most concentrations in all three lines, with the largest separation in SW48 at 1000 nM (~1.6× vs ~3.65×). [CORRECTED] It does not hold universally: in SW480 at 1000 nM mosperafenib exceeds encorafenib in both the monotherapy and the +cetuximab block. n unstated, no statistics, three CRC lines, no skin-relevant model.
Verdict: PARTIALLY-SUPPORTED (downgraded from supported). This is still the strongest genuinely mosperafenib-specific paradox evidence in the public record, and it establishes attenuation, not abolition, and not uniform attenuation.

Claim 3 — "proves paradox breaking properties by not triggering MAPK over-activation in BRAF WT contexts AT ALL LEVELS"

Venue: conference — AACR 2025 Abstract 5606. Cancer Res 2025;85(8_Suppl_1). DOI 10.1158/1538-7445.AM2025-5606. Abstract text retrieved verbatim via Crossref.
Evidence: No cell lines, concentrations or denominators are given for the paradox assertion; it is stated as a conclusion. Judged against the same programme's later readable data, the absolute quantifier "at all levels" is directly falsified (pERK rises ~2.7-3.65× in HCT116).
Verdict: CONTRADICTED. The single most overstated formulation found anywhere — and the abstract that also carries the Compound Ia identity bridge.

Claim 4 — "did not induce RAF paradoxical activation" (Compound Ia)

Venue: peer — Clin Cancer Res 2022;28(4):770-780. DOI 10.1158/1078-0432.CCR-21-2761. PMID 34782366.
Who: Wichmann J, Rynn C, Friess T, ... Dummer R, Levesque MP, Schnetzler G, Martoglio B, Bischoff JR, Pettazzoni P.
Evidence: THE FOUNDATIONAL CLAIM — evidence could not be verified. Bronze-OA PDF at aacrjournals.org returns HTTP 403 (Cloudflare challenge page confirmed by inspecting the downloaded bytes); no Europe PMC full-text XML. The abstract, retrieved verbatim, names no cell line, no concentration and no assay for the paradox experiment; Experimental Design says only "characterized in vitro, ex vivo, and in several preclinical in vivo models of melanoma." The paper never uses the names mosperafenib / RG6344 / RO7276389.
Verdict: UNKNOWN. Highest-value paid-access target in the file — it is simultaneously the origin of the paradox designation, the cited source of the vendor affinity figures, and the cited source of the review's "experimentally confirmed BRAF/BRAF dimer" entry.

Claim 5 — "None of the typical BRAFi class toxicities... have been observed to date, highlighting the paradox breaking properties"

Venue: conference — ASCO 2025 Abstract 3542. J Clin Oncol 2025;43(16_suppl):3542. Data cut 25 Sep 2024. Abstract retrieved verbatim via Crossref.
[CORRECTED] Who: Fontana E (first author), Vieito Villar M, Castanon Alvarez E, Matos I, Bechter OE, Moreno I, Han CH, Eefsen RL, Pinato DJ, Plummer R, Prenen H + Roche (Schnetzler, Pettazzoni, Dejardin, Abiraj, Cinato, Flinn, Kratochwil, Roller, Keshelava). No author named Kotani exists on this record.
Evidence: n=51 mCRC (27 / 53% prior BRAFi), monotherapy escalation. The factual half — zero cSCC, PPE or keratoacanthoma — is a real single-arm observation. The inferential half — "highlighting the paradox breaking properties" — is an unsupported causal leap and is the sharpest claim/evidence divergence in the file. Absent: any on-treatment skin biopsy or tissue pERK; any comparator arm; adequate n or exposure. [CORRECTED] The abstract's pharmacodynamic sentence is "reaching Ctrough levels exceeding pERK inhibition > 80%" — it does not say "PK-derived"; that qualifier belongs to CT017 and the JCO paper. The inference that it is modelled rather than measured is sound but was quoted as if stated.
Verdict: ASSERTED-ONLY.

Claim 6 — same absence statement, causal clause removed

Venue: conference — AACR 2025 Abstract CT017. Cancer Res 2025;85(8_Suppl_2). DOI 10.1158/1538-7445.AM2025-CT017. Data cut 5 Dec 2024. Retrieved verbatim.
Evidence: n=66 monotherapy (56% prior BRAFi; 11 melanoma, 51 CRC, 4 other) + 12 in the cobimetinib combination (not reported). Grade 3 TRAEs 11/66 (16.6%); grade 4 in 2 (3%), both laboratory; no grade 5; 2 (3%) discontinued for TRAE; one DLT of grade 3 rash maculo-papular. ORR 25% in 64 evaluable. PD stated as "Ctrough levels exceeding PK-derived pERK inhibition >90%."
[CORRECTED] Verdict: ASSERTED-ONLY as a paradox claim (downgraded from partially-supported). The descriptive observation stands as reported fact and is recorded as such in §6; the mechanism is not established by it, and the title's "paradox breaker" designation is imported, not demonstrated here.

Claim 7 — JCO phase I: title designation "paradox breaker"; "no reports of PPE or keratoacanthoma"

Venue: peer — J Clin Oncol 2026;44(14):1337-1348. DOI 10.1200/JCO-25-02444. PMID 41894647. Published 10 May 2026. Trial ISRCTN13713551. Abstract retrieved verbatim via Europe PMC.
Evidence — MATERIAL FINDING, RE-VERIFIED AND CONFIRMED. n=80 (63 CRC, 13 melanoma, 4 other; 60% BRAFi-exposed), median treatment 3.7 months (0.2-28.6). The claim shrank on the way to peer review. Both conference abstracts (n=51, n=66) asserted that cutaneous squamous cell carcinoma had not been observed; the peer-reviewed paper at n=80 states only "There were no reports of palmar-plantar erythrodysesthesia or keratoacanthoma." cSCC is not mentioned. Whether an event occurred, or the sentence was narrowed to what the data support, cannot be determined — full text is Cloudflare-blocked. The paper presents no paradoxical-activation experiment of its own; the designation is imported from Wichmann 2022. PD is again modelled: "sustained PK-derived pERK inhibition ≥90%." Two dose-limiting toxicities, both cutaneous: grade 3 rash and grade 3 rash maculopapular.
Verdict: ASSERTED-ONLY.

Claim 8 — "Mosperafenib is a novel, paradox-breaking BRAF inhibitor"

Venue: peer — Cancer Res Commun 2026;6(6):1435-1446. DOI 10.1158/2767-9764.CRC-26-0196. PMID 42127914. PMC13276731 (open access; full text retrieved and read).
Evidence: None. Verified against the full text: the descriptor appears in the abstract opening line and once in the introduction ("designed to overcome this paradox, showing promising activity in preclinical models (6)", ref 6 = Wichmann 2022). The string "pERK" does not occur anywhere in the paper. No wild-type or RAS-mutant model, no paradox experiment, no skin or tissue PD. It is a ctDNA biomarker analysis (49 biomarker-evaluable mCRC patients).
Verdict: ASSERTED-ONLY. Recorded because propagation count is itself a CI signal: naive evidence-counting scores three peer-reviewed "paradox breaker" papers when only one contains a relevant experiment.

Claim 9 — [CORRECTED — THIS ENTRY IS NOT ABOUT MOSPERAFENIB]

Claim as written: compounds of the claimed formula "show considerably less paradoxial [sic] activation of the MAPK signalling pathway while retaining high potency."
Venue: company (patent) — WO2021116050A1, "New BRAF inhibitors as paradox breakers." F. Hoffmann-La Roche. Inventors Dolente, Hewings, Hunziker, Krummenacher, Pettazzoni, Wichmann. Priority 2019-12-10; published 2021-06-17. Full text retrieved and searched.
[CORRECTED] Claim 1 of this patent EXPRESSLY EXCLUDES MOSPERAFENIB, verbatim: "...with the proviso that (3R)-N-[2-cyano-4-fluoro-3-(3-methyl-4-oxo-quinazolin-6-yl)oxy-phenyl]-3-fluoro-pyrrolidine-1-sulfonamide is excluded or a pharmaceutically acceptable salt thereof." That structure is mosperafenib. The proviso repeats in claims 1, 2 and 3 and throughout the embodiment paragraphs. The prior audit's caveat — that Example 3 is the 4-chloro analogue and the mapping is "unresolved" — is now resolved in the negative: the molecule is not merely unexemplified here, it is affirmatively disclaimed, which is the standard signature of a compound claimed in an earlier family member.
[FLAG] The parent/priority family that actually claims mosperafenib was not identified. A patent-family search (INPADOC, on the excluded structure or on CAS 2649372-20-1) is the next step, and that document — not this one — would carry any per-compound paradox data for this molecule.
[CORRECTED] Assay details, for the series (not this molecule): HTRF phospho-ERK (Thr202/Tyr204) run in parallel in A375 (BRAF V600E, inhibition readout) and HCT116 (BRAF wild-type / KRAS-mutant, paradoxical activation readout); Cisbio Advanced ERK phospho-T202/Y204 kit; cells from ATCC, banked by the Roche repository. The comparator is DABRAFENIB (plus reference compound AR-25 from WO2012/118492), not encorafenib. The quantified conclusion is: "the maximum paradox inducing effect is substantially reduced by more than 25% for all examples" — i.e. a ≥25% reduction versus dabrafenib, meaning every exemplified compound still induces a measurable paradox signal. EC50 is declared "not applicable" where activation stays below 50% of the dabrafenib maximum.
Verdict: UNKNOWN as to mosperafenib. The patent's honest relative phrasing remains a useful contrast with the sponsor's absolute journal/congress phrasing, but it is a statement about a chemical series from which this molecule is carved out, and it must not be presented as the sponsor's claim about mosperafenib.

Claim 10 — review classification of "Compound Ia"

Venue: review — Acta Pharm Sin B 2024, "Targeting RAF dimers in RAS mutant tumors: From biology to clinic." PMC11120325. Full text retrieved and table decoded.
Content: Classifies Compound Ia among "Paradox breakers" (with PLX8394, PLX7904). Conformation NA. Cell-free IC50 (nmol/L): ARAF NA, BRAF <5E-10, CRAF NA, BRAF V600E 1.77E-09. Dimer column (headed "confirmed experimentally"): BRAF/BRAF. Scope "Effective in BRAF mutants in melanoma". Status Preclinical. Refs 146/147 = Bonfill-Teixidor 2022, Wichmann 2022. Argues that paradox breakers, as αC-helix-out inhibitors with negative allostery, have limited potency in class III BRAF and RAS-mutant tumours, and states that "compound Ia... and hybrid 6... neither shows potent activity against RAS mutant tumors."
Evidence: Secondary throughout; the review generated no data. Its "not potent against RAS-mutant tumours" statement is an efficacy statement and must not be conflated with a paradox statement.
Verdict: UNKNOWN. Awareness only; cannot establish or refute paradoxical activation.

Claim 11 — [NEW — MISSED BY THE PRIOR AUDIT] review asserts RAF-dimerisation inhibition

Claim: "Conventional BRAF inhibitors may paradoxically activate MAPK signaling in RAS-mutant or WT cells via RAF dimerization, contributing to toxicity and resistance. In contrast, 'paradox-breaker' BRAF inhibitors (e.g., PLX8394, PLX7904, mosperafenib) inhibit RAF dimerization and maintain MAPK suppression."
Venue: review — Biomolecules 2026;16:543. PMC13114179. DOI 10.3390/biom16040543. Independent (non-Roche) authors. Full text retrieved.
Evidence: None. This is the only source located that makes an explicit RAF-dimerisation claim about mosperafenib by name, and it is a review sentence with no supporting experiment, grouping the molecule with the PLX series by assumption. It is a live contamination route into CI files and LLM summaries.
Verdict: ASSERTED-ONLY.

Claim 12 — absence of claim in Roche's patient-facing channel (recorded as a finding)

Venue: company — forpatients.roche.com trial page for RO7276389. Trial status Completed.
Content: Describes RO7276389 only as "an experimental medicine" for BRAF-mutated tumours. No paradox, paradox-breaker, RAF-dimer or mechanism claim of any kind. No investor deck, press release or pipeline page located makes the paradox claim.
Reading: This is the opposite of the usual CI pattern. The overstatement here occurs in the scientific literature, not in the marketing.
Verdict: UNKNOWN (null result, informative).

Claim 13 — vendor catalogue copy

Venue: company (vendor) — MedKoo 59708; Selleck RG6344; MedChemExpress "B-Raf IN 2"; ProbeChem PC-72243.
Content: "potent, selective, brain penetrant BRAF inhibitor with binding Kd of 0.6, 1.2, and ~1.7 nM for BRAF WT, BRAF V600E, and c-RAF, presenting paradox breaker properties"; plus kinase IC50 <1.77 nM across BRAF mutants and cytotoxicity 5.2-30.2 nM across 94 cell lines.
[CORRECTED] Evidence: The ProbeChem datasheet does cite a primary source — Wichmann J, et al. Clin Cancer Res. 2022;28(4):770-780. The prior audit's "no primary citation traceable" is wrong. The figures remain awareness only (the cited primary is unreadable), but they are now corroborated in pattern by an independent review's table (§3), which strengthens rather than weakens the pan-RAF-binding flag.
Verdict: ASSERTED-ONLY. Must never be cited as grounding.

Claim 14 — independent peer-reviewed pushback (not a paradox claim; closed out)

Venue: peer — Zhao C, Luo R, Yan H, Li Y. "Prior BRAF Inhibitor Exposure Is Not Enough to Define Resistance in the Phase I Mosperafenib Study." J Clin Oncol 2026. DOI 10.1200/jco-26-00996, PMID 42659591. With Reply: Vieito M, Fontana E, Schnetzler G, Dejardin D, Renner F, Roller A, Keshelava N. DOI 10.1200/jco-26-01636, PMID 42659589. Both 27 Aug 2026.
Evidence: Confirmed via Europe PMC that neither item has an abstract (abstractText = NONE for both); full texts are Cloudflare-blocked. Content assessed from titles only.
Verdict: UNKNOWN. Cannot be excluded that the exchange touches safety or paradox. Only independent peer-reviewed challenge to any mosperafenib claim; worth paid retrieval.

6. Behaviour by BRAF alteration

Monomeric BRAF V600E — the only context with substantial data. Cellular pERK IC50 ~10.8 nM (LS411N) and ~9.79 nM (HT29); IC80 31.87 / 25.14; IC95 100.1 / 72.57. Encorafenib is marginally more potent at IC50 and IC80 in both lines (8.13 / 6.3 and 29.18 / 19.37) but marginally less potent at IC95 in LS411N (105.5 vs 100.1). Mosperafenib is not differentiated on V600E potency; its claimed differentiation is tolerability and achievable target coverage. In vivo, monotherapy outperformed encorafenib + cetuximab at clinically relevant doses in BRAFi-naive LS411N and HT29 xenografts; combinations with cetuximab and with FOLFOX gave regressions, including in PDX from patients progressing on encorafenib/cetuximab. Clinically: ORR 24.2% across 80 patients (2 CR, 14 PR); mPFS 6.4 months CRC, 3.5 months melanoma; ORR 25% BRAFi-naive vs 14.8% BRAFi-experienced mCRC at the ASCO cut (mPFS 7.3 vs 3.6 months). Brain-penetrant.

RAS-mutant context — preclinical only, and the data cut against the strong claim. HCT116 and SW480 (both BRAF WT / KRAS mut): mosperafenib raises pERK, up to ~3.65× DMSO (HCT116, 100 nM + cetuximab) and ~2.85× in monotherapy, returning toward baseline at 1000 nM monotherapy. Encorafenib raises it more at most points, but not all (SW480 at 1000 nM). Paradoxical activation in RAS-mutant cells is attenuated but present, at concentrations spanning the compound's own cellular IC50-IC95 window against V600E. No NRAS- or HRAS-mutant model. No RAS-mutant in vivo work. No clinical data whatsoever — phase I eligibility was BRAF V600-mutant only. A review states Compound Ia does not show potent activity against RAS-mutant tumours (review, efficacy not paradox).

BRAF WT / RAS WT. SW48: ~1.15 / 1.45 / 1.6× at 10/100/1000 nM monotherapy — mild but non-zero, monotonically increasing with dose. In the +cetuximab block all values fall below the overall DMSO control.

[NEW] BRAF V600E under prolonged exposure. Suppl Fig S3: P-ERK rebounds at 24-48 h and P-c-RAF is induced under mosperafenib 300 nM in HT29; cetuximab blunts both. [FLAG] Read from blot intensity without densitometry — human verification required. This is EGFR-feedback reactivation, not classical wild-type dimer paradox; do not conflate, but do not omit either.

Dimeric BRAF fusions (KIAA1549-BRAF) and splice variants (p61) — UNKNOWN. No potency, cellular, in vivo or clinical data. Conspicuous, since this is the context in which the PLX-series paradox breakers were specifically characterised.

Class II BRAF (RAS-independent dimerising: K601E, G469A, G464V, fusions) — UNKNOWN. No data.

Class III BRAF (kinase-impaired, RAS-dependent: D594G, G466V, N581S) — UNKNOWN. No data. The Acta review predicts a class-level limitation for αC-out paradox breakers, but that is a class-level inference in a review, and this molecule's conformational class is itself undisclosed.

Acquired resistance. MAPK reactivation under drug is the dominant route. In the melanoma brain-metastasis work (compound 1a / C1a — [FLAG] identity to Compound Ia inferred, not stated), BRAF kinase-domain duplication was a frequent driver. Clinically, preexisting MAPK-pathway alterations in BRAFi-experienced patients correlated with lack of response (HR 3.5, P=0.003); BRAFi-naive patients acquired them at progression; molecular response at C1D15 in 100% of BRAFi-naive vs 48% of BRAFi-experienced (Cancer Res Commun 2026, n=49 biomarker-evaluable).

7. Clinical paradox signal, and development status

All clinical data derive from one trial: ISRCTN13713551, phase Ia/b, open-label, single-arm, Roche-sponsored, BRAF V600-mutant solid tumours only. No randomised comparator exists anywhere, so every cutaneous safety statement below is an uncontrolled observation.

The three data cuts, and how the claim changed across them

	ASCO 3542 (`conference`)	AACR CT017 (`conference`)	JCO 2026 (`peer`)
Data cut	25 Sep 2024	5 Dec 2024	final
n (monotherapy)	51 mCRC	66 (+12 cobimetinib, unreported)	80
Prior BRAFi	53%	56%	60%
cSCC asserted absent	yes	yes	NOT MENTIONED
PPE / keratoacanthoma absent	yes	yes	yes
Causal "paradox breaking" clause	yes	no	no (title only)
Grade 3(-4) TRAEs	8/51 (14.5%)	11/66 (16.6%)	13/80 (16.3%)
Grade 4	2 (3.6%), laboratory	2 (3%), laboratory	—
Discontinued for TRAE	3 (5.5%)	2 (3%)	2 (2.5%)
DLTs	—	1 × G3 rash maculo-papular	2, both cutaneous: G3 rash, G3 rash maculopapular
PD statement	"Ctrough exceeding pERK inhibition >80%"	"PK-derived pERK inhibition >90%"	"sustained PK-derived pERK inhibition ≥90%"

Reading (re-verified, unchanged). The strongest and most-quoted clinical proof-point for paradox breaking — absence of cuSCC — survives only in the two conference abstracts. It does not appear in the peer-reviewed publication at the larger denominator. Whether an event occurred between the December 2024 cut and the final analysis, or the claim was simply narrowed, cannot be established. Any CI file citing "no cuSCC" for mosperafenib is citing conference material, not peer-reviewed material, and must say so.

Signals consistent with attenuated (not absent) paradox. - Cutaneous toxicity was dose-limiting. Both DLTs in the final n=80 were grade 3 rash / rash maculopapular. Rash is not diagnostic of paradoxical MAPK activation, but it is the tissue and the toxicity class in question, and it is what capped the dose. MTD not reached to 3,600 mg/day. - Consistent with the preclinical picture: ~1.5-3.65× pERK induction in BRAF wild-type cells — attenuation, not abolition. - Roche's own patent background explicitly names "accelerated growth of secondary tumours upon BRAFi monotherapy (mainly keratochantoma [sic] and squamous-cell carcinomas)" as the clinical consequence of the paradox — the sponsor's own framing of what absence would have to demonstrate.

Not reported / denominator caveats. - New primary melanoma: no rate at any cut, in any venue. Not stated as absent; simply not addressed. - RAS-mutant tumour progression: untestable from this trial. Eligibility was BRAF V600-mutant only; no RAS-mutant patient was ever dosed. The canonical clinical paradox safety signal has never been tested for this molecule. - Second primary malignancies, verrucous keratoses, hyperkeratosis, actinic keratosis, alopecia, panniculitis: no rates published. - Exposure duration: median 3.7 months. cuSCC/KA on first-generation BRAFi typically emerges within the first several months — marginally adequate for a first look, far from definitive at n=80 single-arm. - Prior BRAFi exposure in 60% is a confounder in both directions, unadjusted in any published cutaneous analysis. - PHARMACODYNAMICS ARE MODELLED, NOT MEASURED. Every clinical pERK figure is PK-derived. No skin biopsy, no paired tumour biopsy, no keratinocyte or normal-tissue pERK is reported anywhere. There is zero direct human pharmacodynamic evidence, of any kind, that mosperafenib fails to activate MAPK in wild-type tissue. The bridge from "no keratoacanthoma in 80 patients" to "paradox breaker" is entirely inferential.

[CORRECTED] [NEW] Development status — the recorded unknown is refuted

company (sponsor pipeline disclosure) — Roche Group development pipeline, Q2 2026 update, status as of 23 July 2026 (https://assets.roche.com/f/176343/x/cb875526bd/pharmahy26.pdf, page 2, "Changes to the development pipeline"):

Removed from phase I — 4 NMEs: RG6344 mosperafenib (BRAFi) – solid tumors; RG6382 CD19 x CD3 – SLE; RG6496 HTT SNP ASO – Huntington's disease; RG6468 NME – solid tumors. 1 AI: RG6418 selnoflast (NLRP3i) – inflammation.

Verified by coordinate-level text extraction: the mosperafenib entry sits at x=49.3, y=164.2, directly beneath the "Removed from phase I" column header at x=89.4, y=209.1. RG6344 appears nowhere else in the 68-page document and is absent from every current phase table — consistent with removal rather than advancement.

Discipline on this finding. Roche states the fact of removal and gives no reason. Do not infer that paradox behaviour, cutaneous toxicity, or the efficacy data caused it. Portfolio prioritisation, competitive positioning and strategic reasons are all equally consistent with the disclosure as written.

[FLAG] A human should confirm against the Roche H1 2026 results presentation and the next quarterly pipeline update, and should check whether any out-licensing or partnering followed.

Verdict on the clinical signal

The observed cutaneous safety profile is compatible with reduced paradoxical activation and was a genuine, if early, differentiator. It does not and cannot establish the mechanism. Single-arm, n=80, median 3.7 months, no tissue PD, no RAS-mutant exposure, cutaneous toxicity dose-limiting, the cuSCC element of the claim absent from the peer-reviewed report — and the asset removed from the sponsor's phase I pipeline as of July 2026.

8. Citations

LABEL TIER — EMPTY. openFDA drug label API returns NOT_FOUND for openfda.generic_name:"mosperafenib", for free-text mosperafenib, and for RG6344 (all queried 1 Sep 2026). No FDA USPI. [FLAG] EMA not conclusively retrieved — record as not retrieved, not as searched and absent. No label-grounded paradox claim exists for this molecule.

PEER 1. Renner F, Eckmann J, Handl C, Wichmann J, Schnetzler G, Kratochwil NA, Keshelava N, Ros J, Gonzalez E, Verges J, Chicote I, Martinez-Quintanilla J, Palmer HG, Pettazzoni P. Mosperafenib, a Novel Paradox-Breaker BRAF Inhibitor with Potent Preclinical Activity in BRAF-Mutated Colorectal Cancer. Mol Cancer Ther 2026;25(4):599-609. DOI 10.1158/1535-7163.MCT-25-0562. PMID 41340484. Abstract via Europe PMC (resultType=core); full text paywalled/Cloudflare-blocked (mct_ft.xml returned 0 bytes). 2. SUPPLEMENTARY DATA, READ DIRECTLY. AACR figshare collection 8398399; 10 items (S1-S9 + Table 1) enumerated via api.figshare.com. Suppl Fig S1 (31925011) — pERK dose-response, HT29 and LS411N, BRAF V600E only, no wild-type arm. Suppl Fig S2 (31924972) — densitometry of Fig 3A/B/C in SW48, HCT116, SW480 at 10/100/1000 nM ± cetuximab 5 µg/mL; both blocks read, at magnification. Suppl Fig S3 (31924942) — HT29 time course ± EGF ± cetuximab; P-ERK, P-c-RAF, P-AKT. These three figures are the primary evidence on which the contradicted and downgraded verdicts rest. Local copies: C:\\Users\\Owner\\AppData\\Local\\Temp\\claude\\C--Users-Owner-Documents-Hugh-Context-Folder-Fore\\06918b60-6b30-474c-84c4-6907c80d5eef\\scratchpad\\mct-25-0562_suppl_fig_s1_suppsf1.png, ...\\mct-25-0562_suppl_fig_s2_suppsf2.png, ...\\mct-25-0562_suppl_fig_s3_suppsf3.png, plus magnified crops ...\\hct116_zoom.png and ...\\sw480_zoom.png. 3. Vieito M, Fontana E, Han CH, Castanon E, Pinato DJ, Bechter O, Eefsen RL, Moreno I, Prenen H, Dummer R, Plummer R + Roche. Phase I Study of Mosperafenib... J Clin Oncol 2026;44(14):1337-1348. DOI 10.1200/JCO-25-02444. PMID 41894647. Abstract retrieved verbatim via Europe PMC; full text Cloudflare-blocked (jco_ft.xml = 0 bytes). 4. Serrano-Serrano ML, Godfried Sie C, Bechter O, Pretelli G, Vieito M, Fontana E et al. Early ctDNA Dynamics Predict Response to Mosperafenib... Cancer Res Commun 2026;6(6):1435-1446. DOI 10.1158/2767-9764.CRC-26-0196. PMID 42127914. PMC13276731. Full text retrieved and read (ctdna_ft.xml); reference list decoded; "pERK" occurs 0 times. 5. FOUNDATIONAL, UNVERIFIABLE — Wichmann J, Rynn C, Friess T, et al. Preclinical Characterization of a Next-Generation Brain Permeable, Paradox Breaker BRAF Inhibitor. Clin Cancer Res 2022;28(4):770-780. DOI 10.1158/1078-0432.CCR-21-2761. PMID 34782366. Compound = "Compound Ia" throughout. Full text NOT obtainable: the downloaded PDF is a Cloudflare challenge page ("Just a moment..."), not the article; no Europe PMC full-text XML. Abstract retrieved verbatim. 6. Bonfill-Teixidor E, Iurlaro R, Handl C, Wichmann J, ... Bischoff JR, Pettazzoni P, Seoane J. Activity and Resistance of a Brain-Permeable Paradox Breaker BRAF Inhibitor in Melanoma Brain Metastasis. Cancer Res 2022;82(14):2552-2564. DOI 10.1158/0008-5472.CAN-21-4152. PMID 35584009. Compound = "compound 1a (C1a)". Abstract retrieved verbatim; full text paywalled. [FLAG] C1a = Compound Ia is an inference. 7. INDEPENDENT PUSHBACK — Zhao C, Luo R, Yan H, Li Y. J Clin Oncol 2026, DOI 10.1200/jco-26-00996, PMID 42659591; Reply DOI 10.1200/jco-26-01636, PMID 42659589; both 27 Aug 2026. Confirmed: neither has an abstract in Europe PMC; both Cloudflare-blocked. Not read.

CONFERENCE (all three abstracts retrieved verbatim via the Crossref JATS abstract field) 8. AACR 2025 Abstract 5606. Cancer Res 2025;85(8_Suppl_1). DOI 10.1158/1538-7445.AM2025-5606. Renner F et al. Sole public source of "RG6344, previously identified as Compound Ia in Wichmann et al., Clin Cancer Res 2022" and of "...not triggering MAPK over-activation in BRAF WT contexts at all levels." 9. AACR 2025 Abstract CT017. Cancer Res 2025;85(8_Suppl_2). DOI 10.1158/1538-7445.AM2025-CT017. Vieito M et al. Data cut 5 Dec 2024, n=66. 10. ASCO 2025 Abstract 3542. J Clin Oncol 2025;43(16_suppl):3542. DOI 10.1200/JCO.2025.43.16_suppl.3542. Fontana E et al. (author list corrected). Data cut 25 Sep 2024, n=51. Source of "...highlighting the paradox breaking properties of this BRAF inhibitor." 11. NOT READ — ESMO-family abstract 126P, Ann Oncol 2026, S0923-7534(26)00311-X. annalsofoncology.org returned HTTP 403. Likely overlaps the JCO dataset.

COMPANY 12. PATENT — WO2021116050A1, "New BRAF inhibitors as paradox breakers." F. Hoffmann-La Roche. Priority 2019-12-10; published 2021-06-17. Full text retrieved and searched. Claim 1 expressly excludes mosperafenib by proviso. Discloses the HTRF pERK A375 / HCT116 assay, dabrafenib and AR-25 comparators, and the ">25% reduction in maximum paradox inducing effect" conclusion for its own examples. https://patents.google.com/patent/WO2021116050A1/en 13. ROCHE PIPELINE — Roche Group development pipeline, pharma half-year 2026, Q2 2026 update, status as of 23 July 2026. Page 2 lists RG6344 mosperafenib (BRAFi) – solid tumors under "Removed from phase I." https://assets.roche.com/f/176343/x/cb875526bd/pharmahy26.pdf 14. REGISTRY — ISRCTN13713551, retrieved via the ISRCTN WHO-format API. Primary sponsor "Roche (United States)"; support F. Hoffmann-La Roche; registered 03/12/2021; target 292; phase IA/B open-label multicentre dose-escalation + cohort expansion of RO7276389 alone and with cobimetinib; status "No longer recruiting". Not on ClinicalTrials.gov (0 studies for mosperafenib, RO7276389 and RG6344). 15. Roche patient-facing trial page for RO7276389 (forpatients.roche.com). Status Completed. Notable null result: describes it only as "an experimental medicine"; makes no mechanism claim. 16. VENDOR — AWARENESS ONLY. ProbeChem RG6344 (https://www.probechem.com/products_RG6344.html) — cites Wichmann 2022 as its primary source; Kd 0.6 / 1.2 / 1.7 nM (BRAF WT / V600E / c-RAF), kinase IC50 <1.77 nM across BRAF mutants, cytotoxicity 5.2-30.2 nM across 94 cell lines; CAS 2649372-20-1, C20H17F2N5O4S, MW 461.444. MedKoo 59708 (HTTP 403 on direct fetch), Selleck RG6344, MedChemExpress "B-Raf IN 2". Must never be cited as grounding. 17. NOT ASSESSED — AdisInsight drug record 800066493 (authentication required); BioWorld article 730587 (HTTP 403); PatSnap Synapse record; DrugHunter "compound Ia" page (does not state the RG6344 equivalence).

REVIEW — AWARENESS ONLY 18. Targeting RAF dimers in RAS mutant tumors: From biology to clinic. Acta Pharm Sin B 2024. PMC11120325. Full text retrieved; table columns decoded (§3). 19. [NEW] A New Era of Salvage-Line Treatment for Metastatic Colorectal Cancer... Biomolecules 2026;16:543. PMC13114179. DOI 10.3390/biom16040543. Independent authors. Asserts mosperafenib "inhibit[s] RAF dimerization" with no supporting data — see claim 11. 20. Novel RAF-directed approaches to overcome current clinical limits and block the RAS/RAF node. Mol Oncol 2024. PMC11161739. Full text keyword-scanned: no mention of RG6344, RO7276389, mosperafenib or Compound Ia. Recorded as a negative search result.

SEARCH COVERAGE / NEGATIVE RESULTS 21. Europe PMC REST API, mosperafenib, resultType=core: 6 records — [CORRECTED] composition is 3 named-molecule primary papers (MCT 2026, JCO 2026, Cancer Res Commun 2026) + the letter/reply pair + 1 review (Biomolecules 2026), not "4 primary papers plus a letter/reply pair." Wichmann 2022 and Bonfill-Teixidor 2022 do not contain the string "mosperafenib" and are not in this set. No independent (non-Roche-affiliated) group has published any paradoxical-activation experiment on this molecule.

9. Unknowns (revised)

BINDING MODE ENTIRELY UNDISCLOSED. No type assignment, DFG/αC/R506 status, contact residues, co-crystal or PDB ID. A specialist review left the field blank while populating it for every comparator paradox breaker — which shows those authors could not source it, not that no assignment exists anywhere.
MONOMER-VERSUS-DIMER POTENCY NEVER PUBLISHED IN READABLE FORM. No dimer-disruption assay, forced-dimer comparison or selectivity ratio is publicly readable. [CORRECTED] A review's dimer column, headed "confirmed experimentally", enters BRAF/BRAF for Compound Ia and cites the two paywalled 2022 papers — so such data may exist behind the paywall.
THE FOUNDATIONAL PARADOX EXPERIMENT IS NOT PUBLICLY VERIFIABLE. Wichmann 2022 is the origin of "did not induce RAF paradoxical activation", is cited for the paradox property by both the MCT 2026 and JCO 2026 papers, is the vendor datasheets' cited source for the affinity figures, and is the review's cited source for the BRAF/BRAF dimer entry — and its full text is unobtainable. Single highest-value paid-access target in this file.
[CORRECTED] IDENTITY. mosperafenib = RG6344 = RO7276389 is peer-reviewed (MCT 2026 abstract). Only the Compound Ia link is conference-only (AACR 5606), and it is partly corroborated at peer tier by Cancer Res Commun 2026 citing Wichmann 2022 as mosperafenib's preclinical characterization. [FLAG] Compound Ia = compound 1a (C1a) remains an unsourced inference.
cSCC STATUS IN THE FINAL DATASET IS UNRESOLVED. Two conference abstracts assert absence; the peer-reviewed paper at n=80 is silent on cSCC. JCO full text Cloudflare-blocked.
NEW PRIMARY MELANOMA RATE NEVER REPORTED at any cut, in any venue.
NO HUMAN PHARMACODYNAMIC EVIDENCE OF ANY KIND FOR THE MECHANISM. All clinical pERK figures are PK-derived. No skin biopsy, no paired tumour biopsy, no normal-tissue pERK anywhere.
NO CLINICAL DATA IN RAS-MUTANT DISEASE, AND NONE POSSIBLE FROM THIS TRIAL. Eligibility was BRAF V600-mutant only.
NO DATA IN DIMERIC BRAF FUSIONS, SPLICE VARIANTS (p61), CLASS II OR CLASS III BRAF.
NO NRAS- OR HRAS-MUTANT MODEL, NO MELANOCYTE, NO KERATINOCYTE OR SKIN MODEL. The published wild-type panel is three colorectal lines.
[CORRECTED] PATENT MAPPING — RESOLVED, NEGATIVE. WO2021116050A1 expressly disclaims mosperafenib in claim 1. [FLAG] The earlier family member that actually claims it was not identified; an INPADOC family search on the excluded structure or CAS 2649372-20-1 is the next step, and that document would carry any per-compound paradox assay values.
[CORRECTED] VENDOR AFFINITIES ARE SOURCED BUT UNVERIFIABLE. ProbeChem cites Wichmann 2022; the Acta review's table is concordant in pattern. Both are secondary and the cited primary is unreadable. If genuine, they imply a compound binding wild-type BRAF at least as tightly as V600E — a profile nobody has reconciled with the paradox-breaker positioning.
FIGURE READ-OFFS ARE APPROXIMATE. All S2 fold-changes were read from bar heights; the numeric data are unpublished and the caption states no n, no replicate structure and no statistics. Direction and rough magnitude are unambiguous; precise values are approximate. The S3 blot read is qualitative. [FLAG]
THE JCO LETTER/REPLY EXCHANGE WAS NOT READ. Confirmed to have no abstract in Europe PMC; both Cloudflare-blocked. It cannot be excluded that safety or paradox is discussed. Only independent peer-reviewed challenge on record.
ESMO-FAMILY ABSTRACT 126P NOT READ — HTTP 403.
[CORRECTED — NO LONGER UNKNOWN] DEVELOPMENT STATUS. Roche's Q2 2026 pipeline update (23 July 2026) lists RG6344 mosperafenib under "Removed from phase I." No reason is disclosed — do not infer one. [FLAG] Confirm against the H1 2026 results presentation and the next quarterly update; check for any subsequent out-licensing.
NO INDEPENDENT REPLICATION EXISTS. Every paradoxical-activation experiment on this molecule in the public record was performed or co-authored by Roche.

10. Bottom line

The paradox-breaker designation for mosperafenib is carried by one unreadable 2022 preclinical paper, propagated into three peer-reviewed venues and three congress abstracts without new mechanistic evidence, and asserted in absolute form ("does not trigger", "at all levels", "did not induce") that the programme's own readable supplementary data contradict — pERK rises up to ~3.65-fold over control in a BRAF wild-type, KRAS-mutant line at concentrations inside the compound's own therapeutic window, and mosperafenib is not uniformly below encorafenib.

The sponsor's patent language is more honest than its journal language, but that patent is not about this molecule — it disclaims it. The single most-quoted clinical proof-point, absence of cuSCC, survives only in conference material and is absent from the peer-reviewed report, in which the two dose-limiting toxicities were both cutaneous. There is no human pharmacodynamic evidence of the mechanism at all, and the compound has never been given to a RAS-mutant patient.

And as of the sponsor's own July 2026 disclosure, the asset has been removed from Roche's phase I pipeline — for reasons Roche does not state.

Data cut: 1 September 2026.

NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026

Verification confidence: high

NXP200 (HSK42360 / HSK42360-Na) — Paradox-Claim Audit, CORRECTED

Data cut: 1 September 2026. Scope: public-domain evidence only. Plixorafenib is out of scope and is not referenced, inferred to, or compared against anywhere in this document.

AUDITOR'S NOTE ON THE PRIOR VERSION. The audit under review was built on an incomplete retrieval that it reported as complete. It missed two clinical conference abstracts (ASCO 2025 #3109; SNO 2025 CTNI-28), mischaracterised the content of AACR 2026 LB405, asserted class II absence that the record refutes, and contained one fabricated finding (a "drafting error" in the Nuvectis press release that does not exist). Its top-line conclusions on label tier, peer tier, binding mode and structure survive and are now verified by retrieval rather than by failed search. Its characterisation of the strength of the sponsor's mechanistic claim does not survive.

1. Identity — RESOLVED (one term-of-deal correction)

Field	Value	Tier
Originator code	HSK42360 (tablet, free form); HSK42360-Na (sodium salt, "second-generation salt form")	conference / registry
Aliases	HSK-42360, NXP-200	review (aggregator)
Originator	Haisco Pharmaceutical Group Co., Ltd. (Chengdu and Shanghai, China)	conference (author affiliations)
Ex-China sponsor	Nuvectis Pharma, Inc. (NASDAQ: NVCT)	company
Licence, 22 June 2026	"exclusive worldwide Ex-China rights"; $40M upfront/near-term, up to $1.421bn milestones, tiered royalties	company
Trial sponsor	Haisco, for all four registered studies	registry (verified via CTG API)
INN	None assigned — none found in any public source	—
Stage	Phase 1 / 1b (China). US IND expected 4Q2026.	company / registry
Approval	None in any jurisdiction	verified absence

CORRECTION — territory. The prior audit said "outside Greater China." The sponsor's own text says "exclusive worldwide Ex-China rights." "Greater China" appears only in Chinese trade coverage and aggregators (review tier). Whether "China" here includes Hong Kong, Macau or Taiwan is not disclosed. Separately, Haisco retains India and certain Southeast Asia territories — for NXP100, not NXP200. [FLAG: confirm the defined "Territory" against the executed licence agreement or the Nuvectis 8-K exhibit before any competitive-geography conclusion. The 8-K exhibit 99.1 (accession 0001104659-26-090094) returned HTTP 403 and remains unaudited.]

Structure: NOT DISCLOSED — verified by retrieval, not assumed. PubChem PUG-REST name lookup returned No CID found for both HSK42360 and NXP200. ChEMBL molecule search returned 0 records. No Haisco patent could be tied to the compound. No chemotype is public.

2. Venue integrity — the tag audit

Nothing in this record is label tier. Nothing in this record is peer tier. Both absences are now verified by retrieval:

Label: openFDA label API queried for HSK42360 and NXP200 → NOT_FOUND for both. No approval by FDA, EMA or NMPA. No USPI, no SmPC, no section number can be cited. (The absence was retrieved and read, not merely searched for.)
Peer: Europe PMC returns exactly one record — Cancer Innovation 2025, "Re-Innovation in Clinical Trial Designs Based on Precision Therapy", DOI 10.1002/cai2.70028, pubType article-commentary — a passing mention with no characterisation data. Review tier. No primary research article on this molecule exists. (The prior audit reported this search as returning nothing. It returned one item.)

⚠️ THE TRAP IN THIS RECORD

Three of the five substantive sources carry journal names and journal DOIs — Cancer Research, Journal of Clinical Oncology, Neuro-Oncology — and are NOT peer-reviewed primary literature. They are supplement-published conference abstracts. Anyone re-tagging 10.1200/jco.2025... or 10.1093/neuonc/noaf201... as "peer" because of the journal name commits precisely the venue-integrity error this audit exists to prevent. Every substantive claim about this molecule sits at conference tier or below.

The only label-tier document invoked anywhere in the NXP200 record is the dabrafenib US Prescribing Information (2025), cited by the sponsor inside SNO 2025 CTNI-28 as a historical efficacy benchmark. It is a label for a different molecule and establishes nothing about NXP200.

Trade press and aggregators (BioWorld, HMP Global, Synapse/PatSnap, BioSpace, Quiver, pharmaceutical-technology) are mapped to review tier — awareness only. Note precisely: these are trade press and database aggregation, not review articles. They restate sponsor copy and add no independent data.

3. The complete evidence base — five substantive sources

CORRECTION: the prior audit had two of these. Crossref returns four works plus one Europe PMC commentary.

#	Source	Venue	Content
1	AACR 2024 Abstract 598 — Gao S, Li Y, Wang J, Tang P, Shi Z, Li Y, Yan P (all Haisco). Cancer Res 2024;84(6_Suppl):598. DOI 10.1158/1538-7445.am2024-598. Meeting 5–10 Apr 2024, San Diego	conference	The only paradox-specific experiment in the record. Preclinical.
2	ASCO 2025 Abstract 3109 — Li J (Beijing Cancer Hospital), Xu T, Zhang W, Li W, Kang Z, Yuan Z, et al. J Clin Oncol 2025;43(16_suppl):3109. DOI 10.1200/jco.2025.43.16_suppl.3109	conference	MISSED BY PRIOR AUDIT. First clinical cut, n=17, DCO 15 Jan 2025.
3	SNO 2025 Abstract CTNI-28 — Li W (Beijing Tiantan), Kang Z, Shen L, Li S, Huang M, Zhang W, Mo L, Guo H, Zhang W. Neuro-Oncology 2025;27(Suppl). DOI 10.1093/neuonc/noaf201.0525	conference	MISSED BY PRIOR AUDIT. Second clinical cut, n=38, DCO 23 May 2025.
4	AACR 2026 Abstract LB405 — Gao S, Qian M, Wang J, Li Y, Jiang S, Wu Q, Li F, Tang P, Shi Z, Li Y, Yan P (all Haisco). Cancer Res 2026;86(8_Suppl):LB405. DOI 10.1158/1538-7445.am2026-lb405. Meeting 17–22 Apr 2026, San Diego, late-breaking	conference	Third clinical cut, n=69, DCO 15 Nov 2025 — plus new preclinical data the prior audit said it did not contain.
5	Nuvectis + Haisco press releases, pipeline page, Q2 2026 results	company	Mechanism and positioning claims. No citations on any of them.

Registry (verified, ClinicalTrials.gov v2 API, 1 Sep 2026): NCT06536400 (Ph1, n=316, start 25 Jun 2024) · NCT07158710 (Ph1, paediatric 6–<18y brain tumours, n=159, start 15 Aug 2025) · NCT07561554 (Ph1, HSK42360-Na, n=159, start 6 Mar 2026) · NCT07790510 (Ph1/2, HSK42360-Na + cetuximab ± chemo, mCRC, n=180, start 12 Aug 2026). All Haisco-sponsored, all recruiting. All four eligibility criteria confirmed to require documented BRAF V600 mutation positivity. A registry-wide search for HSK42360 returns these four and no others; no NXP200-named study is registered. China CDE: CTR20242136, CTR20262192.

4. Binding mode — NOT DISCLOSED

No public source classifies NXP200/HSK42360 as type I, type I½, or type II. No DFG-in/DFG-out or αC configuration is stated anywhere. No contact residues. No co-crystal structure. No PDB ID. No SPR/ITC or any biophysical data. There is no structural biology of any kind in the public record.

All mechanism statements are functional, not structural — and they are not equivalent to one another. The escalation chain, corrected and now complete:

Rung	Wording	Venue	What it would take to establish it
A	"did not induce homodimer or heterodimer of CRAF with CRAF or BRAF" (HCT116)	conference — AACR 2024 #598	This is the experiment. Narrow but real.
B	"displays significantly less paradoxical activation than approved BRAF inhibitors and spares wt-BRAF-containing RAF dimers"	conference — ASCO 2025 #3109 and SNO 2025 CTNI-28	MISSED BY PRIOR AUDIT. A relative, unquantified claim. "Spares wt-BRAF-containing dimers" is a distinct structural-selectivity assertion with no supporting data shown.
C	"effectively inhibits signaling and disrupts BRAF homo- and hetero-dimers, with significantly reduced paradoxical activation"	conference — AACR 2026 LB405	"Disrupts" is a stronger verb than rung A ("failure to induce") supports. Note the same sentence keeps "reduced", not "eliminated".
D	"circumvents this by physically disrupting the formation of the BRAF dimer"	company — Nuvectis pipeline page, no citations	Active physical disruption. No structural or biophysical data of any kind exist to support it.

The critical finding the prior audit missed sits at rung B. In every venue where the sponsor addresses scientific peers, the claim is reduced paradoxical activation — comparative, relative, and unquantified. In company-tier material it becomes circumvents — absolute. No public source quantifies the reduction. No source states that paradoxical activation is absent.

Whether the molecule achieves this by monomer selectivity or by dimer binding cannot be determined from the public record.

5. Potency and monomer-vs-dimer — LARGELY NOT DISCLOSED

One potency value exists in the entire record: "HSK42360 is a potent BRAF^V600E inhibitor with IC50 value of 5 nM at enzymatic level" (AACR 2024 #598).

CORRECTION: the prior audit attached the wild-type selectivity claim to the enzymatic assay. It is made at the cellular level and separately: HSK42360 "exhibited significant anti-proliferation activity against multiple tumor cell lines with BRAF^V600E... but was highly selective for BRAF wild-type cells." No fold-selectivity number is given in either place.

Still not disclosed: monomer-vs-dimer potency pair · BRAF/CRAF heterodimer enzymatic IC50 · CRAF IC50 · any potency value against class II or class III mutants or BRAF fusions · kinase selectivity panel · any biochemical dimer-affinity measurement.

CORRECTION — dimer-driven models. The prior audit stated there is "no cell-based potency against a dimer-driven model." Overstated. Two dimer/resistance-driven in vivo models exist, neither yielding a potency value: - NRASmA375 CDX (NRAS Q61K, BRAF V600E) — "first gen-resistant" model. The abstract's own word is "Surprisingly, HSK42360 demonstrated promising antitumor effects" — the sponsor flags this as not predicted by its own model. Worth recording: retained activity in a RAF-dimer-driven setting sits in tension with a purely monomer-selective account, and the sponsor does not reconcile the two. - Vemurafenib + cetuximab–resistant CRC model (LB405) — "HSK42360 monotherapy outperformed the combination therapy" and showed synergy with trametinib. MISSED BY PRIOR AUDIT.

Dimer detection method: still NOT NAMED. No co-IP, crosslinking, BRET or NanoBiT is stated in any abstract. No dose-response, no quantification, no pERK immunoblot is shown for the dimerisation experiment.

Brain penetration — CORRECTED and expanded. AACR 2024: "Kp,uu greater than 1". LB405 gives species values: Kp,uu 1.09 (mouse), 2.8 (rat) — described by the sponsor as "best-in-class brain penetration potential" (an uncontrolled superlative; no comparator values given).

6. Behaviour by alteration class

Monomeric BRAF V600E — the only alteration with real data

Preclinical (conference): enzymatic IC50 5 nM; anti-proliferation in A375 (melanoma), DU4475 (breast), COLO205 (colon); tumour activity in COLO205 and A375 CDX; melanoma brain micro-metastasis model with BLI imaging, prolonged survival. Glioma PDX TGI 67%, 108%, 111.5% across three dose groups, no significant body-weight change (LB405 — missed by prior audit). Preclinical PD — CORRECTION: the prior audit stated there is no pERK data anywhere. AACR 2024 #598: "Immunohistochemistry showed a close correlation between the inhibition of tumor growth and ERK1/2 phosphorylation in A375 CDX tumor tissues." In vivo tumour pERK IHC exists.

Clinical (conference, three sequential cuts of NCT06536400):

Cut	DCO	n	Efficacy
ASCO 2025 #3109	15 Jan 2025	17	ORR 18.2% (2/11 evaluable) — 1 CRC PR, 1 ganglioglioma PR, 3 SD with shrinkage (RECIST or RANO)
SNO 2025 CTNI-28	23 May 2025	38 (9 PCNST)	PCNST ORR 62.5% (5/8 evaluable: 4/7 HGG PR, 1/1 LGG PR, 2 SD)
AACR 2026 LB405	15 Nov 2025	69 (23 PCNST)	PCNST ORR 40.9% (9/22, 1 CR); HGG 36.8% (7/19, "including preliminary PR"); LGG 66.7% (2/3)

[FLAG: the PCNST ORR moves 62.5% (5/8) → 40.9% (9/22) as the evaluable denominator grows. This is an ordinary small-denominator effect, but the company-tier ">40% response rate" framing is anchored to the larger, lower number while the earlier, higher number was itself promoted at SNO. A human should decide how to characterise the trajectory.] [FLAG: the HGG 36.8% includes unconfirmed ("preliminary") partial responses — LB405's own wording. Only one response in the entire dataset is described as confirmed (the single CR). The prior audit reported these percentages without the caveat.]

Missed comparative claim: SNO 2025 CTNI-28 makes a naive cross-trial comparison inside the abstract — "An approved BRAF V600E inhibitor combined with a MEK inhibitor, had ORRs of 33% and 50% in BRAF V600E-mutated HGG and LGG, respectively (dabrafenib US Prescribing Information, 2025)." Different trial, different population, different era, no adjustment. Awareness only.

Dimeric BRAF fusions — UNKNOWN

No fusion construct, no fusion-driven cell line, no fusion-bearing patient, no fusion potency value in any source, preclinical or clinical.

Class II (RAS-independent, dimerising) — ASSERTED AT CONFERENCE TIER, NO DATA SHOWN

CORRECTION — the prior audit's absolute negative is false. ASCO 2025 #3109 and SNO 2025 CTNI-28 both state: "Treatment with HSK42360 results in excellent and durable anti-tumor effect in BRAF Class I and II mutant CDX or PDX models." A conference-tier class II activity claim exists. It remains evidence-free: no class II mutant is named (no K601E, L597, G469A, G464V), no model is identified, no tumour volume, TGI or potency figure is given, and the claim appears only in background/framing sentences. All four registered trials require BRAF V600 for entry, so no clinical class II evidence can exist at this cut.

Class III (RAS-dependent, kinase-impaired) — UNKNOWN

Genuinely absent from every source. No class III mutant is named or tested anywhere, preclinical or clinical. The Nuvectis pipeline page nonetheless claims inhibition "of the dimerizing, RAS independent (Class II) and RAS-dependent (Class III) mutations." That is company-tier, uncited, and unsupported by any retrievable experiment.

RAS-mutant / BRAF wild-type — one experiment, narrow

HCT116 (KRAS G13D, BRAF wild-type) is the single cell line in which the paradox claim was tested. It remains the only such experiment in the record after full re-retrieval. The NRASmA375 model is a BRAF V600E melanoma that acquired NRAS Q61K — a resistance model, not a BRAF-wild-type RAS-driven tumour — and therefore does not test whether the drug is inert in RAS-mutant wild-type-BRAF tissue. No efficacy claim is made in RAS-mutant BRAF-wild-type tumours. No RAS-mutant or non-V600 patient is eligible for any registered trial.

7. Clinical paradox signal — NO READOUT EITHER WAY (finding, not a clean bill of health)

Across all three clinical abstracts (n=17 → 38 → 69, DCOs Jan 2025 → May 2025 → Nov 2025):

Cut	Safety reported
ASCO 2025	64.7% TEAEs; most frequent increased ALT 23.5%, increased AST 23.5%; 89.7% of TEAEs grade 1; 2 drug-related grade 3 (creatinine, ALT); 1 drug-related SAE (creatinine); no DLT, no grade 4, no treatment-related discontinuation or death
SNO 2025	81.6% TEAEs, 68.4% drug-related; 79.6% of TRAEs grade 1; grade 3 TRAEs 7.9%; 2 drug-related SAEs (grade 3 creatinine, grade 2 pyrexia); TRAE incidence in PCNST comparable to overall; no DLT, no grade 4, no treatment-related discontinuation or death
AACR 2026 LB405	81.2% drug-related TEAEs; increased ALT 23.2%, increased AST 23.2%, anaemia 20.3%; most TEAEs grade 1; no DLTs, no treatment-related discontinuations or deaths

Not one of the three abstracts mentions cutaneous squamous cell carcinoma, keratoacanthoma, hyperkeratosis, skin papilloma, rash, palmar-plantar erythrodysaesthesia, or any new primary malignancy — in either direction. No abstract reports such events; no abstract states that none occurred. The prior audit reached this conclusion from one abstract; it now holds across three, which strengthens it as a silence finding without converting it into evidence of absence.

Absence of mention in a dose-escalation abstract is not evidence of absence. Undisclosed in all three: median follow-up, exposure duration, dermatological surveillance protocol, denominator-time per dose level. The retrievable fact bearing on exposure is that NCT06536400 started 25 June 2024 and the LB405 data cutoff is 15 November 2025 — roughly 17 months of trial operation, with per-patient exposure unknown. (Reasoning, not evidence: paradox-driven cuSCC and keratoacanthoma with first-generation agents characteristically emerge over months of exposure in heavily sun-damaged melanoma populations, whereas this cohort is enriched for primary CNS tumours. This shapes the prior on what a null dermatology signal here would mean; it is not data about NXP200 and must not be recorded as such.)

No clinical pharmacodynamic data — no tumour or skin pERK at any dose in any study, despite pharmacodynamics being a stated secondary objective of NCT06536400 (ASCO 2025, Methods).

By design, no RAS-mutant tumour-progression signal is observable: every registered trial requires BRAF V600.

The Nuvectis claim that paradox-breaking will reduce first-generation toxicities is therefore not supported by any comparative clinical dataset: no comparator arm, no cross-trial toxicity comparison with numbers, no dermatology-specific reporting. (Note the narrowing: a cross-trial efficacy comparison with numbers does exist, at SNO 2025, against the dabrafenib USPI.)

Upcoming: "Oral presentation of additional clinical data generated by Haisco will be presented at the upcoming European Society for Medical Oncology (ESMO) conference (Madrid, Spain, October 23–27, 2026)" — not public at this cut, and the most likely near-term source of new paradox-relevant safety data.

8. Paradox claims register — corrected verdicts

#	Claim	Who	Venue	Verdict	Basis
1	"did not induce homodimer or heterodimer of CRAF with CRAF or BRAF and thereby blocked reactivation of the MAPK pathway in HCT116 cells, whereas Dabrafenib significantly induced RAF dimerization and MAPK reactivation"	Gao et al., Haisco	conference (AACR 2024 #598)	partially-supported	This is the experiment — the only paradox-specific data in the record. One BRAF-WT/KRAS G13D line. Assay not named. No dose-response, no quantified blot, no second RAS-mutant line, no keratinocyte/skin model, no in vivo paradox readout. Comparator: dabrafenib only.
2	Title designation: "a potent, brain permeable, BRAF paradox breaker"	Gao et al., Haisco	conference (AACR 2024 #598)	partially-supported	Rests entirely on claim 1, plus 5 nM enzymatic IC50 and an unquantified cellular WT selectivity statement. No monomer-vs-dimer pair, no class II/III or fusion data, no structural evidence.
3	"displays significantly less paradoxical activation than approved BRAF inhibitors and spares wt-BRAF-containing RAF dimers"	Li J / Li W et al. (academic first authors, Haisco-sponsored trials)	conference (ASCO 2025 #3109; SNO 2025 CTNI-28)	asserted-only	ADDED — missed by prior audit. Stated as background in both abstracts. No data shown for either half. "Spares wt-BRAF-containing dimers" is a structural-selectivity claim with zero structural or biochemical support. Note it is relative and unquantified — the honest version of the company's absolute language.
4	"excellent and durable anti-tumor effect in BRAF Class I and II mutant CDX or PDX models"	Li J / Li W et al.	conference (ASCO 2025 #3109; SNO 2025 CTNI-28)	asserted-only	ADDED — missed by prior audit; refutes the prior audit's "no class II anywhere". No class II mutant named, no model named, no TGI or volume data, no figure. Background sentence only.
5	"effectively inhibits signaling and disrupts BRAF homo- and hetero-dimers, with significantly reduced paradoxical activation"	Gao et al., Haisco	conference (AACR 2026 LB405)	asserted-only	BASIS CORRECTED. LB405 does present preclinical results (glioma PDX TGI 67/108/111.5%, Kp,uu 1.09 mouse / 2.8 rat, vemurafenib+cetuximab-resistant CRC model, trametinib synergy) — the prior audit wrongly said it contained none. But no dimerisation method, assay, dose-response, quantification or figure accompanies the mechanistic sentence. "Disrupts" remains a stronger verb than the underlying "failure to induce" data support.
6	"NXP200 circumvents this by physically disrupting the formation of the BRAF dimer, resulting in inhibition of the Class I mutation, and also of the dimerizing, RAS independent (Class II) and RAS-dependent (Class III) mutations"	Nuvectis Pharma	company (pipeline page, accessed 1 Sep 2026, no references or footnotes on the page)	asserted-only	No citation. Active physical disruption is not established by claim 1 (failure to induce ≠ disruption). No structural or biophysical data exist. The class III half has no supporting experiment of any kind in the entire record; the class II half rests on claim 4, which shows no data.
7	"for the treatment of BRAF V600X-mutated and Class II/III non-V600-mutated malignancies"; paradoxical activation "results in drug resistance and secondary malignancies"	Nuvectis Pharma	company (PR 22 Jun 2026; Q2 2026 results 4 Aug 2026)	asserted-only	No class III data anywhere. Class II supported only by an undocumented conference background sentence. "Secondary malignancies" is true of the field but carries no evidence about NXP200 — no cutaneous or second-primary event is reported in any of the three clinical abstracts, in either direction.
8	"paradox breaking properties... including emerging resistance to first generation BRAFi while reducing toxicities associated with their usage"	Nuvectis Pharma	company	asserted-only	No comparative toxicity data of any kind. No comparator arm, no cross-trial toxicity comparison with numbers, no cutaneous reporting. The resistance half is weakly supported by two preclinical resistance models (NRASmA375 CDX; vemurafenib+cetuximab-resistant CRC) plus uncontrolled clinical responses in BRAFi-pretreated patients. The toxicity half is forward-looking. [FLAG: this exact sentence could not be located in GlobeNewswire release 3338215 on retrieval. The prior audit cited it there. Source attribution unverified — a human must locate the original.]
9	"best-in-class BRAF paradoxical breaker inhibitor designed to overcome acquired resistance"	Haisco	company (PR Newswire 302807803, 22–23 Jun 2026)	asserted-only	No head-to-head against any other paradox breaker or type II RAF inhibitor is published. "Best-in-class" has no comparative dataset.
10	Sodium salt: "early data indeed demonstrate a marked improved PK and greater single agent clinical activity"	Nuvectis Pharma	company (PR 22 Jun 2026)	asserted-only	ADDED — missed by prior audit. No PK parameters, no exposure figures, no response numbers, no denominators, no comparison shown. NCT07561554 (HSK42360-Na) started 6 Mar 2026 and has reported nothing publicly.
11	">40% response rate in low- and high-grade adult glioma, including one Complete Response. Durable responses also in NSCLC, colorectal and papillary thyroid cancers"	Nuvectis Pharma	company (PR 22 Jun 2026)	split — DOWNGRADED	Glioma: corroborated by LB405 (40.9%, 9/22, 1 CR) — though the HGG component includes unconfirmed PRs. Colorectal: weakly corroborated — ASCO 2025 reports 1 CRC PR among 11 evaluable at n=17. NSCLC and papillary thyroid: asserted-only — no patient numbers, denominators or durations exist anywhere. "Durable" has no supporting duration figure for any tumour type in any public source. The prior audit graded the whole claim "partially-supported"; that let the unsupported components ride on the supported one.
12	"HSK-42360, a BRAF paradox breaker with efficacy in models of brain metastasis and BRAF-inhibitor resistance"	BioWorld (trade press)	review	asserted-only — DOWNGRADED	Secondary reporting of AACR 2024 #598; no independent data. Full text not retrievable (HTTP 403); only the headline could be verified. The prior audit graded this "partially-supported" — a headline that could not be read past cannot support anything. Awareness only.
13	"best-in-class BRAF paradoxical breaker inhibitor that can inhibit BRAF V600X mutations while blocking BRAF dimer formation..."	Synapse/PatSnap; Chinese trade aggregators	review	asserted-only	Restates sponsor copy. Database aggregation. Awareness only — cannot establish mechanism.

9. Unknowns — corrected

Binding mode entirely unknown. No type I / I½ / II designation, no DFG or αC configuration, no contact residues, no co-crystal, no PDB ID, no SPR/ITC data.
Chemical structure not disclosed — verified: PubChem No CID found, ChEMBL 0 records, no attributable Haisco patent.
Monomer-vs-dimer potency not disclosed. No paired IC50s, no CRAF or heterodimer values. Sole potency figure: BRAF V600E enzymatic IC50 = 5 nM.
Fold-selectivity over wild-type BRAF asserted but never quantified (and asserted at the cellular, not enzymatic, level).
The magnitude of "significantly less/reduced paradoxical activation" is nowhere quantified — the sponsor's central peer-facing claim has no number attached to it in any source. (Added.)
The dimerisation assay is not named in AACR 2024 #598; no method, dose-response, quantification or blot is public. LB405's "disrupts dimers" statement likewise carries no method or data.
No paradox testing in any second RAS-mutant line, in BRAF-wild-type keratinocytes/skin models, or in vivo. HCT116 remains a single-line, single-comparator (dabrafenib) observation after full re-retrieval.
Zero data on class III mutants or BRAF fusions, preclinical or clinical. Class II is asserted at conference tier ("Class I and II mutant CDX or PDX models") with no mutant named and no data shown — an asserted claim, not an absence. (Corrected.)
No clinical pharmacodynamic data (tumour or skin pERK) at any dose in any study, despite PD being a stated secondary objective. (Preclinical tumour pERK IHC does exist — AACR 2024 #598.)
Cutaneous SCC, keratoacanthoma, hyperkeratosis, skin papilloma and new primary malignancy are neither reported nor excluded in any of the three clinical abstracts. Median follow-up, exposure duration and dermatological surveillance methodology are all undisclosed. LB405 data cutoff 15 Nov 2025 against a 25 Jun 2024 trial start.
No RAS-mutant tumour-progression signal is observable by design — all four trials require BRAF V600.
Full abstract bodies of AACR #598 and LB405 could not be read on the AACR site (HTTP 403); text was obtained from the Crossref-deposited full abstract, which is the publisher's own deposit and is quoted verbatim throughout this document. ASCO #3109 and SNO CTNI-28 were likewise read from publisher Crossref deposits. No poster or slide deck for any of the four abstracts is in hand. [FLAG: figures, blots, assay methods and full AE tables live in the posters/slides, not the abstracts. Every "method not named" and "no data shown" finding here is a statement about the abstract text only.]
NSCLC and papillary thyroid "durable responses" have no supporting numbers, denominators or durations anywhere.
The Nuvectis 8-K exhibit 99.1 (accession 0001104659-26-090094) returned HTTP 403; corporate-deck paradox language in it is unaudited. Larvol's LB405 page also returned 403.
ESMO 2026 (Madrid, 23–27 Oct 2026) oral presentation of additional Haisco clinical data — not public at this cut; the most likely near-term source of paradox-relevant safety data.
No INN assigned. No US IND filed (expected 4Q2026). No regulatory review document of any kind exists to audit in any jurisdiction.
All preclinical and all paradox-specific data originate from Haisco employees. The clinical abstracts (ASCO 2025, SNO 2025) are first-authored by academic investigators at Beijing Cancer Hospital and Beijing Tiantan Hospital with no Haisco author listed — but these report sponsor-run trials and constitute no independent verification of the mechanism. No independent laboratory has published any characterisation of this molecule. (Corrected from the prior audit's flat "every datum originates from Haisco employees".)

10. Bottom line

The paradox-breaker designation for NXP200 rests, after complete re-retrieval, on one narrow experiment — HCT116, one comparator, method not named, no quantification — supplemented by two in vivo resistance models and a set of repeated but undocumented background assertions across four conference abstracts. There is no label-tier and no peer-tier evidence for this molecule anywhere, and both absences are verified by retrieval.

The material correction to the prior audit is not that it was too harsh — it is that it graded the wrong version of the claim. The sponsor's own peer-facing language is "significantly less paradoxical activation" and "spares wild-type-BRAF-containing RAF dimers": relative, comparative, and unquantified. The company-tier language converts that into "circumvents" and "physically disrupts." The gap between those two is the finding, and no public experiment closes it.

tovorafenib (OJEMDA) — data cut 1 September 2026

Verification confidence: high

Tovorafenib (OJEMDA) — Paradox-Claim Audit, ADVERSARIALLY CORRECTED

Data cut: 1 September 2026. Plixorafenib (FORE8394 / PLX8394) is out of scope and is not researched, inferred, or compared against anywhere in this document.

Venue tiers used here. label = approved regulatory label only (FDA USPI / EMA SmPC), section number quoted. regulatory-assessment = regulator-authored assessment document that is NOT the label (CHMP EPAR, FDA review) — grounding-quality but a distinct tier, added in this revision. peer = peer-reviewed primary literature. conference = abstract/poster/presentation. company = press release, deck, website. review = review or commentary. Label, regulatory-assessment and peer are grounding; conference is provisional; company and review are awareness only and can never on their own establish that a molecule does or does not cause paradoxical activation.

What changed in this revision

The prior audit tagged the CHMP Assessment Report as label. It is not. That single retag is the most consequential correction here, because the audit's headline finding — a regulator contradicting the anti-paradox claim — was being presented at label tier. The finding survives, verified verbatim, but at regulatory-assessment tier. Neither approved label contains the word "paradoxical." Verified by direct retrieval of both.

Also corrected: US PI 12.1 is two sentences, not one; CHMP reproduced three enzymatic IC50s, not four; a Hill slope that undercuts the cooperativity argument was omitted; the Kilburn citation year is wrong; the FIREFLY-1 Arm 1 denominator is stated wrongly; and one leg of the equipotency support is a structural prediction being passed off as corroboration.

Resolved identity

INN tovorafenib. Aliases DAY101, MLN2480, TAK-580, BIIB-024 — all confirmed — plus AMG-2112819 and BSK1369. CAS 1096708-71-2. UNII ZN90E4027M. Trade name OJEMDA (US and EU).

Chain of title. Originated at Sunesis Pharmaceuticals (now Viracta Therapeutics, Inc.); licensed to Millennium Pharmaceuticals (Takeda affiliate), developed as MLN2480/TAK-580 in adult solid tumours and melanoma. On 16 December 2019, DOT Therapeutics-1, Inc. — a Day One subsidiary, not Day One Biopharmaceuticals, Inc. directly — entered an asset purchase agreement with Millennium for the TAK-580 technology rights and know-how; Takeda's exclusive Sunesis/Viracta licence was assigned to DOT-1 (Day One Form 10-K). Current sponsor Day One Biopharmaceuticals, Inc. [FLAG: confirm current HQ city — the ASCO 2022 poster affiliation reads South San Francisco, CA; recent filings may read Brisbane, CA.]

Ex-US. Exclusive ex-US licence to Ipsen Pharma (Paris), announced 25 July 2024 (~$111M upfront, ~$71M cash + $40M equity; up to ~$350M milestones; tiered double-digit royalties from mid-teens). Ipsen Pharma is the EU marketing authorisation holder.

US. FDA accelerated approval 23 April 2024, NDA 217700 (tablets) / 218033 (oral suspension). Current label revision 8/2025; DailyMed SPL version dated 2 September 2025, setid ea3a9631-3a66-6a7c-e053-2995a90ae2ad. [FLAG: the accessdata.fda.gov PDF path cited in the prior audit returns HTTP 404 — retrieve via DailyMed.]

EU. CHMP positive opinion 26 February 2026 (EMA/67438/2026, EMEA/H/C/006140/0000); European Commission conditional marketing authorisation 20 April 2026; documents published 11 May 2026.

Binding mode

Type II — grounded at label tier in both jurisdictions. US PI 12.1 and EU SmPC 5.1 both say "Type II RAF kinase inhibitor."

DFG-OUT, alphaC-IN. Tkacik et al., J Biol Chem 2023;299(5):104634 (Dana-Farber/Harvard, Eck lab). Co-crystals show the inhibitor spanning the nucleotide-binding site, the DFG phenylalanine flipped out, alphaC in the inward position, and both protomers of a back-to-back kinase-domain dimer occupied with identical binding mode. PDB: 6V34 (BRAF V600E + tovorafenib), 8F7O (WT BRAF + tovorafenib), 8F7P (WT BRAF + naporafenib). Contact residues confirmed: C532, T529, K483, D594, E501. [FLAG: crystal resolutions "2.75–3.5 Å" not verified at source.]

Disclosure qualifier added. M.J.E. holds a Novartis consultancy and sponsored research support from Novartis Institutes for Biomedical Research, plus SpringWorks. Novartis sponsors naporafenib, the head-to-head comparator in that paper. Funding otherwise NIH (R35CA242461, P50CA165962, F32CA247198), PLGA fund of the Pediatric Brain Tumor Foundation, Chleck Family Foundation. Independent of tovorafenib's sponsor — but not comparator-neutral.

CHMP concurs (regulatory-assessment, EPAR p.32, verbatim): "Crystallographic studies showed tovorafenib to be an R-enantiomer that displaces the phenylalanine out of the back-pocket of the active site of BRAF, shifts the orientation of the C-helix, and changes the conformation of the glycine-rich loop. This binding mode is similar to that of sorafenib and differs from that of the Type-I BRAF inhibitor vemurafenib."

Clayton et al. — reclassified. Mechanism of Dimer Selectivity and Binding Cooperativity of BRAF Inhibitors, eLife reviewed preprint 95334 v3 (14 January 2025), DOI 10.7554/eLife.95334.3 — no longer preprint-stage. From PDB 6V34 the alphaC position deviates 0.2 Å and the K–E distance 0.4 Å between protomers, from which the authors predict equipotent rather than dimer-selective behaviour. This is a structure-based prediction, not an experimental corroboration, and the authors themselves say the prediction was confirmed in prior studies — i.e. it loops back to Sun and Tkacik rather than adding an independent line. The prior audit's use of it as the "structural basis for it being classed EQUIPOTENT" is structure→behaviour overreach and has been withdrawn as a support.

Monomer vs dimer

No head-to-head cellular monomer-versus-dimer potency ratio has been disclosed by anyone.

1) Sponsor biochemical panel — Rastogi et al., Cancer Res Commun 2025;5(4):668–679, Table 2 (all six authors Day One employees/stockholders; Day One funded). Enzymatic IC50: BRAF V600E 7.1 nM; WT BRAF 10.1 nM; WT CRAF 0.7 nM; ARAF 55 nM. Comparator TAK-632 ARAF 215 nM. [FLAG: TAK-632 WT BRAF 8.3 nM and CRAF 1.4 nM not verified at source.]

CORRECTION. CHMP did not reproduce the ARAF figure. EPAR p.32 verbatim: "In vitro, tovorafenib inhibited the kinase activity of BRAF V600E, WT BRAF, and WT CRAF kinases, with IC50 values of 7.1, 10.1, and 0.7 nM, respectively. However, activity on ARAF was not presented, but it was reported in the literature that tovorafenib is not potent against ARAF (Tkacik et al., 2023)." Three values reproduced, not four.

2) Independent biochemistry — Tkacik 2023, purified constructs, IC50 / Ki (nM): ARAF-SSDD >3000 / >40; BRAF-WT 633 ± 160 / 6.13 ± 5.33; BRAF V600E (monomeric in this assay) 495 ± 117 / 20.5 ± 12.4; CRAF-WT 94.2 ± 13.2 / 1.03 ± 0.99; CRAF-SSDD 84.5 ± 15.8 / 0.21 ± 0.06. On Ki, ~3.3× more potent on the WT BRAF dimer and ~20–100× more potent on CRAF constructs than on monomeric BRAF V600E — dimer-favouring in this assay, not strictly equipotent.

3) Cooperativity — CORRECTED. Hill slopes as actually reported: ARAF-SSDD −1.12; BRAF-WT −2.89; BRAF V600E −0.75; CRAF-WT −1.55 ± 0.43; CRAF-SSDD −3.18.

CORRECTION. The prior audit omitted CRAF-WT −1.55 and asserted a "−2.6 to −3.2 range for dimers" that appears nowhere in the paper. Only two of the four non-V600E constructs show steep slopes; CRAF-WT sits nearer the monomeric controls. The cooperativity argument — previously called "the strongest primary evidence underpinning the anti-paradox rationale" — is materially weaker than presented, and it remains purified-protein data with no cellular pERK readout in RAS-mutant or BRAF wild-type cells anywhere in that paper. [FLAG: verify the cooperativity sentence verbatim — the retrieved text reads "...binding of inhibitor to the active site of one protomer increases affinity for inhibitor binding to the second protomer in the RAF dimer", which differs slightly from the prior audit's quotation.]

4) Sun et al. 2017 asserts "MLN2480 binds with equal affinity to BRAF monomers and to preformed BRAF dimers." This is an assertion, not a binding measurement. What supports it is matched cellular potency in p53-null murine neuroprogenitors (KIAA1549:BRAF 190 nM vs BRAF V600E 248 nM, per EPAR p.32–33) plus activator/receiver transactivation constructs.

Not disclosed. No cell-based paradox index for tovorafenib. No HaCaT-HRAS-G12V keratinocyte paradox assay for tovorafenib, by anyone.

Behaviour by alteration class

Monomeric BRAF V600E — active, approved. Cellular IC50 248 nM (Sun 2017). Xenograft regression in A375 (V600E) and WM-266-4 (V600D). Both labels cover BRAF V600 pLGG. Tkacik shows no positive cooperativity against the monomer.

Dimeric BRAF fusions — the approved use and the best-evidenced setting. Cellular IC50 190 nM (KIAA1549:BRAF); regression in the AGK::BRAF fusion melanoma PDX. CHMP flags the preclinical thinness verbatim: "Only two BRAF fusion models were presented" (EPAR p.32).

Class II BRAF point mutants — UNKNOWN. No systematic potency dataset published. Fusions are dimer-signalling and sometimes grouped as class II, but the discrete point-mutant panel does not exist publicly. Not in either label.

Class III BRAF — UNKNOWN. A complete evidentiary void. No potency, no pERK, no clinical data. Mechanistically the highest-risk class for paradoxical activation with any RAF inhibitor.

RAS-mutant — mixed, and the weakest link. Rasco et al. 2023: NRAS-mutant melanoma, 31 evaluable (Q2D treatment-naive n=14: 1 PR / 4 SD / 9 PD; QW treatment-naive n=17: 0 PR / 9 SD / 8 PD). No hyperprogression or accelerated progression reported. Tumour pERK fell ≥70% (median) in the Q2D expansion cohorts including the NRAS-naive cohort, but only 12% (H-score) and 8% (image analysis) at day 21 in the QW cohort — weak target engagement on the schedule that is now approved. CHMP (EPAR p.32): "Sensitivity to tovorafenib for Ras mutations and WT B-RAF was variable." [FLAG: the prior audit's "32 NRAS-mutant" total and its "Q2D pre-treated n=1" NRAS patient were not located in the retrieved full text.]

NF1 loss-of-function (RAS-activated, BRAF wild-type) — MAPK ACTIVATION IS DOCUMENTED IN BOTH APPROVED LABELS.

US PI 13.2, verbatim: "In vitro, tovorafenib increased phosphorylation of ERK at clinically relevant concentrations in cells with neurofibromatosis Type 1-loss of function (NF1-LOF) suggesting activation, rather than inhibition, of the MAP kinase pathway. In an NF1 genetically engineered mouse model of plexiform neurofibroma without BRAF alteration, tovorafenib did not have antitumor activity, and while not statistically significant, an increase in tumor volume was noted in 2/12 mice (approximately 17%)."

US PI 5.6, verbatim: "Based on nonclinical data in NF1 models without BRAF alterations, tovorafenib may promote tumor growth in patients with NF1 tumors. Confirm evidence of a BRAF alteration prior to initiation of treatment with OJEMDA."

EU SmPC 4.4 and 5.3 carry materially the same text — verified by direct retrieval of the SmPC PDF.

Neither label uses the word "paradoxical." Both describe the phenomenon.

CORRECTION — the GEMM endpoint is described three different ways. Labels: "increase in tumour volume ... in 2/12 mice." CHMP EPAR p.44: "A subset of larger tumours (n=5) was also observed in treatment group with an increase in tumour number in 2/12 mice (approximately 17%)." Rastogi 2025: "A qualitative difference was observed in nine of 48 proximal nerve volume measurements in the tovorafenib-treated group which were relatively larger, (i.e., >3 mm3)." Volume, number, and 9/48 measurements are not the same endpoint. [FLAG: a human must reconcile which endpoint the 2/12 figure actually refers to.]

BRAF wild-type, non-RAS-activated — EPAR p.32: tovorafenib "had adequate potency against in vivo BRAF WT tumours" (Calu-6, BxPC-3) and "No evidence of paradoxical activation was observed in these studies." In murine neuroprogenitors transduced with HRAS-G12V, paradoxical activation was absent at ~1 µM but present at subtherapeutic 0.01–0.1 µM (Sun 2017).

Paradox claims — corrected register

1. `label` — US PI 12.1 makes no paradox claim and no monomer/dimer claim

Who: FDA-approved USPI. Source: OJEMDA USPI 12.1, NDA 217700/218033, rev. 8/2025, DailyMed setid ea3a9631-3a66-6a7c-e053-2995a90ae2ad.

CORRECTION. Section 12.1 is two sentences, not one. Verbatim: "Tovorafenib is a Type II RAF kinase inhibitor of mutant BRAF V600E, wild-type BRAF, and wild-type CRAF kinases. Tovorafenib exhibited antitumor activity in cultured cells and xenograft tumor models harboring BRAF V600E and V600D mutations, and in a xenograft model harboring a BRAF fusion."

Evidence basis: Re-verified by direct retrieval. "Paradoxical," "dimer," "squamous cell carcinoma" and "keratoacanthoma" are absent from the entire US label — this is verified absence, not a search miss. Warnings 5.1–5.6: Hemorrhage; Skin Toxicity Including Photosensitivity; Hepatotoxicity; Effect on Growth; Embryo-Fetal Toxicity; NF1 Associated Tumors. Section 12.2 contains two subsections — Exposure Response Relationships and Cardiac Electrophysiology — and no MAPK pharmacodynamic claim (prior audit's "only exposure-response" was incomplete). Verdict: supported.

2. `label` — the NF1 ERK-activation text

Who: FDA USPI 13.2 + 5.6; EU SmPC 5.3 + 4.4. Evidence basis: Day One in-house MSD pERK ELISA in sNF96.2 (MPNST), MeWo (melanoma), NCI-H1838 (lung) at 1 h and 6 h; Nf1-flox/flox;Postn-Cre+ plexiform neurofibroma GEMM. Published as Rastogi 2025 and the Bessler 2024 Global NF Conference poster. This is regulator-endorsed, label-tier documentation of MAPK activation by tovorafenib in BRAF wild-type, RAS-pathway-activated cells at clinically relevant concentrations. It materially qualifies any blanket "does not cause paradoxical activation" claim. Verdict: supported.

3. `label` — EU SmPC 5.1 DOES make the monomer-and-dimer claim

Verbatim, verified from the SmPC PDF: "Tovorafenib is a central nervous system (CNS) penetrant, selective small molecule Type II RAF kinase inhibitor of mutant BRAF V600E, wild-type BRAF, and wild-type CRAF kinases, including RAF monomers and dimers and BRAF fusion, suppressing activation of the mitogen-activated protein kinase (MAPK) pathway (see section 5.3)."

"Paradox," "squamous cell carcinoma" and "keratoacanthoma" return zero hits across the SmPC; "dimer" and "monomer" appear only in this sentence. Evidence basis: CHMP grounds the monomer/dimer statement on crystallography plus single-isoform enzymatic IC50s — not on any submitted head-to-head monomer-vs-dimer potency comparison. Note the cross-reference from 5.1 straight to 5.3, the NF1 ERK-activation text: the EU label internally qualifies its own claim. Verdict: partially-supported.

4. `regulatory-assessment` (WAS TAGGED `label` — CORRECTED) — CHMP's own paradox assessment

Who: CHMP rapporteurs. Source: CHMP Assessment Report, Ojemda, EMA/67438/2026, adopted 26 February 2026, p.44/154 — verified verbatim:

"This effect is suggestive of paradoxical activation of the pathway in NF1-LOF tumour cell lines. Indeed, paradoxical activation by tovorafenib at subtherapeutic doses (0.01–0.1 µM) was also described in literature. This could be linked to tovorafenib being ARAF sparing. Therefore, the limited non-clinical data do not support consistent activity of tovorafenib, and paradoxical activation could be an issue in this setting."

And: "Although tumour induction was not statistically significant, in parallel with increase pERK observed in vitro, there is a biologically relevant cause for concern regarding safety of this population."

Evidence basis: CHMP reasons from (i) Day One's bell-shaped pERK data, (ii) Sun 2017's subtherapeutic paradoxical activation, (iii) the ARAF-sparing profile taken from Tkacik 2023 because the applicant did not submit ARAF data, (iv) the 2/12-mouse result.

TIER CORRECTION. This is a regulator's assessment document, not an approved label. The finding stands and is the single most important entry in this register — a regulator, on the sponsor's own data, expressly contradicts the unqualified anti-paradox claim in the RAS-activated setting — but it must not be cited as label text. None of this language reached either approved label. Verdict: supported, at regulatory-assessment tier.

5. `peer` — Sun 2017's dose-conditional finding

Who: Sun Y, Alberta JA, Pilarz C, … Stiles CD (Dana-Farber/Harvard; A Kid's Brain Tumor Cure Foundation and NIH funded; no Takeda/Day One authorship). Source: Neuro Oncol 2017;19(6):774–785, DOI 10.1093/neuonc/now261, PMID 28082416 (verified).

Claim: "Under identical conditions, paradoxical activation is not observed in cells treated with type II inhibitors at therapeutic concentrations." In the same paper: "Both MLN2480 and AZD628 give some paradoxical activation at subtherapeutic doses (0.01–0.1 µM)."

Evidence basis: Immunoblot pERK time-courses in murine neural progenitors transduced with HRAS-G12V vs vemurafenib (~1 µM) and dabrafenib (~0.1 µM); HEK293 activator/receiver BRAF transfection experiments showing MLN2480 blocks dimeric signalling only via the drug-bound receiver protomer — catalytic suppression, not prevention of dimerisation. This is the originating primary evidence for the entire anti-paradox narrative, and it is dose-conditional.

Evidentiary-quality caveat, verified verbatim from the corrigendum (Neuro Oncol 2024;26(5):985): a vinculin loading control was used twice across Figure 5 panels A and C, and panel A was replaced with an independent repeat; and an undeclared method detail — MLN2480 was maintained in borosilicate-coated glass tubes because of "high affinity for plastic that can confound the replication of EC50s in cell culture studies." Balance added: the corrigendum also states that the central conclusions remain valid and have been confirmed by subsequent research, and that the plastic issue does not affect clinical trials or animal studies using API. Independent replication of the HRAS-G12V paradox experiments has not been published. Verdict: partially-supported.

6. `peer` — Sun 2017's equipotency claim

Claim: "MLN2480 binds with equal affinity to BRAF monomers and to preformed BRAF dimers"; "an equipotent antagonist of BRAF V600E, KIAA1549:BRAF, and other noncanonical BRAF oncoproteins that function as dimers."

Evidence basis: not a direct binding measurement to preformed dimers. Supported by matched cellular potency (190 vs 248 nM) and the activator/receiver constructs. The Clayton corroboration is withdrawn — Clayton predicts equipotency from crystal geometry and defers to prior experiment; it is not an independent line. Tkacik's biochemistry is in mild tension with strict equipotency (dimer-favouring on Ki). Verdict: partially-supported.

7. `peer` — the pivotal registrational publication's unqualified claim

Who: Kilburn LB, Khuong-Quang DA, Hansford JR, et al., with Day One employees among the authors; Day One funded. Source (CORRECTED): Nature Medicine 2024;30(1):207–217, published online 17 November 2023. DOI 10.1038/s41591-023-02668-y, PMID 37978284. Author Correction DOI 10.1038/s41591-024-02910-1.

Claim, in the introduction: "In contrast to type I BRAF inhibitors, tovorafenib did not induce paradoxical activation of the MAPK pathway in these models." [FLAG: this sentence and its reference number could not be verified verbatim — the article is paywalled. The prior audit's attribution to "ref 25 = Sun et al. 2017" is unverified and must be checked against the published text.]

Evidence basis: A preclinical claim carried into a clinical paper with no clinical pharmacodynamic support in FIREFLY-1 itself — no skin or tumour pERK data were collected. If the Sun citation is confirmed, the citation drops Sun's own explicit subtherapeutic-dose qualifier. The FIREFLY-1 safety data are consistent with the claim but do not test it. Verdict: partially-supported.

8. `peer` (sponsor-authored) — Rastogi 2025 asserts and undercuts the claim in one paper

Who: Rastogi S, Perino S, Lal-Nag M, Wang Y, Blackman SC, Venetsanakos E — all six Day One employees and/or stockholders; Day One funded. Source: Cancer Res Commun 2025;5(4):668–679, DOI 10.1158/2767-9764.CRC-24-0451, PMID 40111124, published 23 April 2025.

Evidence basis: The anti-paradox half is scoped to BRAF-fusion tumours and supported there (AGK::BRAF PDX regression; pERK suppression). The paradox-positive half is supported by MSD pERK ELISA at 1 h and 6 h in sNF96.2, MeWo and NCI-H1838: verbatim, "Increased pERK levels were observed at lower concentrations of tovorafenib and decreased pERK levels at higher tovorafenib concentrations" — a bell-shaped curve. Vemurafenib, run as comparator, showed a monotonic dose-dependent pERK rise in all three lines. Tovorafenib was inactive in 2 of 3 NF1-LOF lines. Day One attributes the effect to ARAF sparing and ARAF:CRAF heterodimer compensation, and proposes vertical RAF+MEK blockade with pimasertib — Loewe 39.8 / Bliss 19.0 in sNF96.2; Loewe 27.2 / Bliss 29.8 in the ERMS PDX ex vivo (the prior audit reported the ≥10 significance threshold as if it were the result).

The honest reading: the claim is true as scoped (BRAF fusions) and false as generalised (RAS-pathway-activated, BRAF wild-type). [FLAG: verify verbatim the sentence "Tovorafenib does not result in paradoxical activation of the MAPK signaling in tumors harboring BRAF fusions" — retrieval returned it in heading-like form.] Verdict: partially-supported.

9. `peer` — Rasco 2023's absence-of-signal argument (DOWNGRADED)

Who: Rasco DW et al.; sponsored by Millennium Pharmaceuticals, Inc. (Takeda). Source: Cancer Chemother Pharmacol 2023;92(1):15–28, DOI 10.1007/s00280-023-04544-5, PMID 37219686.

Verbatim: "In line with preclinical data suggesting that tovorafenib does not trigger paradoxical activation of ERK signaling, squamous cell carcinoma of the skin was reported as a TEAE in only 1 (< 1%) of 149 treated patients."

Evidence basis: N=149 (Q2D 110, QW 39). Adults with melanoma and other solid tumours — sun-damaged skin, the substrate for RAF-inhibitor cuSCC — including 31 evaluable NRAS-mutant melanoma patients. 1 cuSCC verified. Keratoacanthoma and new primary melanoma are NOT REPORTED — this is absence of report, not a counted zero, and the prior audit's "0/149" overstates it.

Downgrade rationale. The number is verified; the claim built on it is not. It is an uncontrolled absence-of-signal in an open-label phase 1 with no formal dermatological surveillance protocol, no comparator arm and short median exposure — and the QW cohort's 12% / 8% median tumour pERK reduction at day 21 means low target occupancy is an untested alternative explanation for the low event rate. Verdict: supported → PARTIALLY-SUPPORTED.

10. `peer` — Tkacik 2023's cooperativity finding (DOWNGRADED)

Source: J Biol Chem 2023;299(5):104634, DOI 10.1016/j.jbc.2023.104634, PMID 36963492. PDB 6V34, 8F7O, 8F7P.

Framing statement, verbatim: "Left unchecked, paradoxical activation can lead to development of secondary skin malignancies, such as squamous cell carcinomas and keratoacanthomas."

Evidence basis: Purified-protein enzymatic assays, four-parameter fitting. Full Hill-slope set: ARAF-SSDD −1.12; BRAF-WT −2.89; BRAF V600E −0.75; CRAF-WT −1.55 ± 0.43; CRAF-SSDD −3.18. Plus three co-crystals with both dimer protomers occupied. This is the best independent mechanistic underpinning for why a type II agent should not leave an unliganded transactivating partner — but it is biochemical and structural only, with no cellular pERK data in RAS-mutant or BRAF wild-type cells, so it cannot on its own establish cellular or clinical absence of paradox. It also identifies the countervailing liability: IC50 >3000 nM against ARAF, the isoform both the sponsor and CHMP invoke to explain the NF1-LOF pERK rise. Verdict: supported → PARTIALLY-SUPPORTED (the cooperativity generalisation does not survive the omitted CRAF-WT slope).

11. `conference` — the ASCO 2022 sponsor poster, asserted without citation

Who: Landi D, Ziegler DS, … Cornelio I, Blackman SC, Da Costa D, Cox MC (Day One), Kilburn L, Hansford JR. Source: FIREFLY-1 (PNOC 026) trials-in-progress poster, ASCO 2022, Abstract TPS10062; J Clin Oncol 40, 2022 (suppl 16).

Claim, verbatim from the poster: "In contrast to type I BRAF inhibitors, tovorafenib does not induce RAS-dependent paradoxical activation of the MAPK pathway."

Evidence basis — CONFIRMED AT SOURCE, and this holds up. The poster's reference list was extracted in full: 1. Morrison, CSH Perspect Biol 2012 · 2. Yaeger & Corcoran, Cancer Discov 2019 · 3. Ryall, Cancer Cell 2020 · 4. Ryall, Acta Neuropathol Commun 2020 · 5. Rankin, Oncologist 2021 · 6. Sun Y, et al. Neuro Oncol. 2017;19:774 · 7. Wright, Neuro Oncol 2020 · 8. Offer, CTOS 2021. The paradox bullet carries no superscript marker. Reference 6 is attached to the next bullet, which ends "…which signals as a RAS-independent monomer⁶". The claim is also broader than Sun supports and adds a "RAS-dependent" qualifier that no cited experiment tested for tovorafenib in a RAS-mutant system. Verdict: asserted-only.

12. `conference` — the Bessler NF1 GEMM poster (DOWNGRADED)

Who: Bessler W, Perino S, Venetsanakos E, Jiang L, Li X, Mang H, et al. (includes Day One employees). Source: Poster, 2024 Global NF Conference, Brussels, 22 June 2024; cited as CHMP footnote 2 at EMA/67438/2026 p.44.

Downgrade rationale. The poster itself was not retrieved by this audit or its predecessor. It reaches the record only through a regulator's footnote and through Rastogi 2025, and the three documents describe its endpoint inconsistently (tumour volume / tumour number / 9-of-48 nerve measurements). A conference item known only at second hand cannot be verified at source. It remains a paradox-positive conference claim from the sponsor's own laboratory that reached both labels via the regulatory dossier. Verdict: supported → PARTIALLY-SUPPORTED. [FLAG: retrieve the poster.]

13. `company` / `conference` — SNO 2025 three-year follow-up

Source: SNO 30th Annual Meeting, Honolulu, 19–23 November 2025; Day One press release 24 November 2025. Median study duration 40.6 months, data cutoff 6 June 2025.

Evidence basis, corrected. What the release actually discloses: median retreatment duration 9 months (all retreated patients still on therapy at cutoff), median maximum tumour reduction −38.3%. The prior audit's assertion that the release reported "no new safety signals" with an unchanged adverse-reaction profile was NOT found in the retrieved release and has been removed. The cuSCC / keratoacanthoma / secondary-malignancy denominator at the 3-year cutoff is not disclosed; only the 10 May 2024 figures survive. Note also that pLGG is paediatric — minimal UV-damaged skin — so a near-zero cuSCC rate has low discriminating power for paradoxical activation regardless of duration. Verdict: unknown.

14. `regulatory-assessment` (WAS TAGGED `label` — CORRECTED) — FIREFLY-1 SPM adverse event of special interest

Source: CHMP Assessment Report, EMA/67438/2026, p.115/154, verified verbatim:

"As of the clinical cutoff date, no patients had events meeting 'non-hematological malignant tumors' search criteria. No patients had any events of squamous cell carcinoma or keratoacanthomas."

Arms 1+2, N=137 (Arm 1 N=77 + Arm 2 N=60), MedDRA v23.1, data cutoff 10 May 2024.

Evidence basis: A prospectively pre-specified AESI with a defined MedDRA search — methodologically stronger than an incidental TEAE tabulation. But: it is CHMP reporting the sponsor's own analysis (secondhand), in a population with low baseline cuSCC risk, with NF1 patients excluded — so it does not probe the RAS-activated setting where the preclinical signal sits. Verdict: supported, at regulatory-assessment tier.

15. `company` — the "About OJEMDA" boilerplate

Label-derived; makes no paradox claim and no monomer/dimer claim. Notable for what it does not say: Day One's approved-product marketing language is more conservative than its scientific-congress language. The unqualified paradox claim lives in posters, protocols and reviews, not in the boilerplate. Verdict: supported (as a characterisation of the boilerplate; awareness tier, establishes nothing about paradox).

16. `company` — "highly-selective" is contradicted by the sponsor's own dossier

Evidence basis: CHMP EPAR p.44 records that tovorafenib inhibits 14 human kinases at 0.2 µM, below unbound Cmax (Arg, Abl, DDR2, EphA1, EphA2, EphA8, Fyn, Lck, Lyn, MuSK, PTK5, SAPK2a, SAPK2b, plus transformed cKit-V560G and PDGFR-V561D), and that "The choice to determine IC50 for 10 out of 14 kinases is questionable." Also inhibits Adenosine A2A, Adenosine A3 and NMDA/polyamine. The monomer/dimer half of the corporate description traces to Sun 2017. Awareness tier; establishes nothing about paradox. Verdict: partially-supported.

17. `review` — the claim hardening away from the data

Sources: Expert Rev Clin Pharmacol 2024;17(11):999–1008, PMID 39412085, DOI 10.1080/17512433.2024.2418405 — verbatim confirmed: tovorafenib "shows superior central nervous system penetration without the paradoxical activation of the MAPK pathway reported for type I BRAF inhibitors." Also DOI 10.1080/14728214.2024.2312817; DOI 10.1093/noajnl/vdaf007; DOI 10.3389/fonc.2023.1154246.

Evidence basis: no new experiments. These restate the class-level type II rationale and, where sourced at all, trace to Sun 2017 and Kilburn 2024. They present the claim unconditionally, stripping both the concentration qualifier Sun attached to it and the NF1-LOF exception both approved labels now carry. The clearest instance in the record of a claim hardening as it moves away from primary data. Review tier is awareness only and cannot establish the absence of paradoxical activation. Verdict: asserted-only.

18. `peer` — no paradox index exists for tovorafenib

Source: Adelmann CH, Ching G, Du L, Saporito RC, Bansal V, Pence LJ, Liang R, Lee W, Tsai KY. Oncotarget 2016;7(21):30453–30460, DOI 10.18632/oncotarget.8351, PMID 27028853 (MD Anderson).

Verified at full text: the paradox index = pERK activation EC80 in HaCaT keratinocytes stably expressing HRAS-G12V, divided by the IC80 against A375. Reported values: vemurafenib 5.5, dabrafenib 10, encorafenib 50 — the prior audit's figures are correct. MLN2480, TAK-580, DAY101 and tovorafenib do not appear anywhere in the paper — verified, not inferred.

CORRECTION. Table 1 contains a fourth compound beyond the three named above. That compound is out of scope for this audit and is not characterised here. The panel should not be presented as three-compound.

Recorded so no reader imports a tovorafenib paradox index from this source. Verdict: unknown.

Clinical paradox signal

Consistent with low paradoxical activation

Rasco 2023, N=149 adults — cuSCC 1/149 (<1%) as a TEAE; keratoacanthoma and new primary melanoma not reported. The most informative denominator in the record: adults with UV-damaged skin, 31 evaluable NRAS-mutant melanoma patients. No accelerated progression or hyperprogression in the NRAS cohorts. Other expansion-phase cutaneous TEAEs: maculopapular rash 30%, pruritus 16%, periorbital oedema 14%, hair colour change 13%, acneiform dermatitis 12%, face oedema 7%.
FIREFLY-1, N=137, cutoff 10 May 2024 — pre-specified SPM AESI, MedDRA v23.1: no SCC, no keratoacanthoma, no non-haematological malignant tumours (regulatory-assessment). Cutaneous TEAEs: hair colour change 76%, maculopapular rash 44% (grade ≥3 in 8%), dry skin 36%, acneiform dermatitis 31%, paronychia 26%; no life-threatening skin reactions.
SNO 2025, median 40.6 months, cutoff 6 June 2025 — long-term exposure. Cutaneous-malignancy denominator not disclosed; the "no new safety signals" characterisation is unverified.

Against, or materially qualifying

Both approved labels carry an NF1 warning grounded in MAPK activation — US PI 5.6 / EU SmPC 4.4 and US PI 13.2 / EU SmPC 5.3, with a non-significant tumour increase in 2/12 mice in the NF1 GEMM. At regulatory-assessment tier, CHMP called this "a biologically relevant cause for concern regarding safety of this population" and said "paradoxical activation could be an issue in this setting."
Confounders. FIREFLY-1 is paediatric — minimal cumulative UV damage, so a near-zero cuSCC rate has little discriminating power — and excluded NF1 patients, precisely the RAS-activated population where the preclinical signal sits. The approved once-weekly schedule produced only 12% / 8% median tumour pERK reduction at day 21 in the phase 1 QW cohort, so low target occupancy is an alternative explanation for a low event rate. The phase 1 had no dermatological surveillance protocol, no comparator, no skin-biopsy pERK.
No post-marketing paradox signal is established. No published FAERS disproportionality analysis, safety communication, label change or case report links tovorafenib to cuSCC, keratoacanthoma or new primary melanoma as of this data cut. Absence of a published analysis is not evidence of absence of a signal.

Unknowns and not-disclosed

No cell-based paradox index for tovorafenib exists, from any source. Verified absent from Adelmann 2016. No equivalent HRAS-mutant keratinocyte assay published by anyone.
No head-to-head cellular monomer-vs-dimer potency comparison has been disclosed. The EU SmPC 5.1 "monomers and dimers" phrase rests on crystallography and single-isoform enzymatic IC50s.
Class II BRAF point mutants: no potency or pERK dataset. Only two BRAF fusion models were submitted to EMA.
Class III BRAF: complete evidentiary void. Highest theoretical paradox risk; nothing published.
No prospective clinical study in a RAS-mutant population with paired skin or tumour pERK biopsies. The only RAS-mutant exposure is inside the phase 1, with tumour pERK but no skin pharmacodynamics.
The 3-year FIREFLY-1 cuSCC / KA / SPM denominator is not disclosed. Only the 10 May 2024 figures are public.
No safety data published from LOGGIC/FIREFLY-2 that would test cutaneous malignancy over longer treatment-naive exposure. EMA has deferred FIREFLY-2 results until July 2030.
No FAERS disproportionality analysis, PV signal assessment, or case report on tovorafenib and cuSCC/KA/new primary melanoma as of 1 September 2026.
Whether the low-concentration pERK rise occurs at drug levels reached in human skin is not established. The labels say "clinically relevant concentrations"; neither label nor sponsor paper maps that to measured human skin or plasma unbound concentrations on the approved 380 mg/m² weekly schedule. (SmPC 5.2 gives steady-state Cmax 6.9 µg/mL and AUC 508 µg·h/mL; SmPC 5.3 gives an hERG IC50 of 8.9 µM described as 32-fold above the clinical plasma unbound concentration in adults — [FLAG: a human should use these to test whether 0.01–0.1 µM is subtherapeutic in skin, which no public document does].)
The magnitude and fold-change of the NF1-LOF pERK rise is not quantified in the accessible text; described qualitatively, and CHMP notes "The trend of the pERK increase at lower concentrations was different between the NF1-LOF tumour cell lines."
ARAF potency is unreconciled by more than 50-fold. Sponsor 55 nM (Rastogi Table 2) vs Tkacik >3000 nM. ARAF was not submitted to EMA at all — CHMP had to source it from the literature. Since ARAF sparing is the mechanism both the sponsor and CHMP invoke to explain the NF1-LOF pERK rise, this is a material gap in the only proposed mechanism.
Founding-paper quality caveat. Sun 2017 required a 2024 corrigendum (vinculin control duplicated across Figure 5 A and C, panel A replaced; undeclared borosilicate-glass handling because of plastic affinity that "can confound the replication of EC50s"). The authors state the central conclusions remain valid. Independent replication of the HRAS-G12V paradox experiments has not been published.

Gaps in the audit itself (not errors — omissions a human must close)

[FLAG] The FDA regulatory record was never retrieved. The audit reasons at length from the European CHMP assessment while consulting no US regulatory review. Missing: the FDA Multi-Discipline Review for NDA 217700 (Drugs@FDA), and Singh S, Bradford D, Chatterjee S, Li X, et al. "FDA Approval Summary: Tovorafenib for Relapsed or Refractory BRAF-Altered Pediatric Low-Grade Glioma." Clin Cancer Res 2025;31(8):1383–1389, PMID 39808502 — FDA-authored and peer-reviewed. Until these are read, "the US regulator made no paradox statement" is silence, not verified absence. Only the US label has been verified.
[FLAG] The Bessler 2024 poster has never been retrieved at source by anyone in this audit chain.
[FLAG] Kilburn 2024's introduction sentence and its reference number are unverified (paywalled).
[FLAG] Rastogi's verbatim "does not result in paradoxical activation…" sentence is unverified.

Citations — corrected

Source	Tier
OJEMDA USPI §§1, 2.1, 5.6, 12.1, 12.2, 13.2; NDA 217700/218033; rev. 8/2025; DailyMed SPL setid ea3a9631-3a66-6a7c-e053-2995a90ae2ad (version dated 2 Sep 2025) — retrieved and read in full	label
Ojemda EU SmPC §§4.4, 5.1, 5.2, 5.3; EMEA/H/C/006140; EC conditional MA 20 Apr 2026; MAH Ipsen Pharma — PDF retrieved, text extracted, keyword-searched	label
CHMP Assessment Report, Ojemda, EMA/67438/2026, adopted 26 Feb 2026, EMEA/H/C/006140/0000 — pp.32–33, p.44, p.115 — PDF retrieved, text extracted	regulatory-assessment (was mis-tagged `label`)
⚠️ FDA Multi-Discipline Review, NDA 217700 (Drugs@FDA) — NOT RETRIEVED	regulatory-assessment
⚠️ Singh S, Bradford D, Chatterjee S, Li X, et al. FDA Approval Summary. Clin Cancer Res 2025;31(8):1383–1389, PMID 39808502 — NOT RETRIEVED, MISSING FROM PRIOR AUDIT	peer (FDA-authored)
Sun Y, Alberta JA, et al. Neuro Oncol 2017;19(6):774–785. DOI 10.1093/neuonc/now261, PMID 28082416	peer
Corrigendum, Neuro Oncol 2024;26(5):985 — retrieved and read	peer
Tkacik E, Li K, Gonzalez-Del Pino G, Park E, Eck MJ, et al. J Biol Chem 2023;299(5):104634. DOI 10.1016/j.jbc.2023.104634, PMID 36963492. PDB 6V34, 8F7O, 8F7P. NIH-funded; Eck: Novartis consultancy and Novartis sponsored research support, SpringWorks	peer
Rastogi S, Perino S, Lal-Nag M, Wang Y, Blackman SC, Venetsanakos E. Cancer Res Commun 2025;5(4):668–679. DOI 10.1158/2767-9764.CRC-24-0451, PMID 40111124. All authors Day One employees/stockholders; Day One funded	peer (sponsor-authored)
Kilburn LB, Khuong-Quang DA, Hansford JR, et al. Nat Med 2024;30(1):207–217 (online 17 Nov 2023). DOI 10.1038/s41591-023-02668-y, PMID 37978284. Correction DOI 10.1038/s41591-024-02910-1. Day One funded	peer (sponsor-funded)
Rasco DW, et al. Cancer Chemother Pharmacol 2023;92(1):15–28. DOI 10.1007/s00280-023-04544-5, PMID 37219686. Millennium/Takeda sponsored	peer (sponsor-funded)
Clayton J, Romany A, Matenoglou E, Gavathiotis E, Poulikakos PI, Shen J. Mechanism of Dimer Selectivity and Binding Cooperativity of BRAF Inhibitors. eLife 95334 v3, 14 Jan 2025. DOI 10.7554/eLife.95334.3. Structure-based prediction, not experimental corroboration	peer
Adelmann CH, Ching G, Du L, Saporito RC, Bansal V, Pence LJ, Liang R, Lee W, Tsai KY. Oncotarget 2016;7(21):30453–30460. DOI 10.18632/oncotarget.8351, PMID 27028853. Verified NOT to contain tovorafenib/MLN2480/TAK-580/DAY101	peer
Landi D, Ziegler DS, … Blackman SC, Cox MC, Kilburn L, Hansford JR. FIREFLY-1 (PNOC 026) TiP poster, ASCO 2022, Abstract TPS10062; J Clin Oncol 40, 2022 (suppl 16) — retrieved; reference list extracted in full	conference
Bessler W, Perino S, Venetsanakos E, Jiang L, Li X, Mang H, et al. Poster, 2024 Global NF Conference, Brussels, 22 Jun 2024. CHMP footnote 2, EMA/67438/2026 p.44 — NOT RETRIEVED AT SOURCE	conference
Day One. Three Year Follow-Up Data From OJEMDA FIREFLY-1, SNO 2025, 24 Nov 2025. Median study duration 40.6 months; cutoff 6 Jun 2025	company / conference
Day One. FDA Accelerated Approval press release, 23 Apr 2024 — source of the "About OJEMDA" boilerplate	company
Day One. FIRELIGHT-1 first patients dosed, 23 May 2022	company
Ipsen / Day One. Exclusive ex-US licensing agreement, 25 July 2024	company
Day One Biopharmaceuticals, Inc. Form 10-K (SEC EDGAR CIK 0001845337) — DOT Therapeutics-1, Inc. asset purchase from Millennium, 16 Dec 2019; assignment of Takeda's exclusive Sunesis (now Viracta Therapeutics) licence	company
Expert Rev Clin Pharmacol 2024;17(11):999–1008, PMID 39412085, DOI 10.1080/17512433.2024.2418405	review
Expert Opin Emerg Drugs 2024, DOI 10.1080/14728214.2024.2312817	review
Neuro-Oncol Adv 2025, DOI 10.1093/noajnl/vdaf007	review
Front Oncol 2023;13:1154246, DOI 10.3389/fonc.2023.1154246	review
Tovorafenib: First Approval. Drugs 2024, DOI 10.1007/s40265-024-02069-6, PMID 38967715. Full text not retrieved	review

Bottom line after adversarial review. The prior audit's central substantive judgement survives and is if anything strengthened by verification: the anti-paradox claim is true as scoped to BRAF fusions and false as generalised to RAS-pathway-activated, BRAF wild-type settings, and both approved labels now carry NF1 language describing MAPK activation without using the word "paradoxical." What does not survive is some of the scaffolding: a regulator's assessment report was being cited as label text, a purified-protein cooperativity argument was presented without the one Hill slope that weakens it, a structural prediction was presented as independent corroboration, and a US regulatory record that would test the "no US paradox statement" claim was never opened.

Vemurafenib (ZELBORAF). Data cut: 1 September 2026.

Verification confidence: high

Vemurafenib (ZELBORAF) — Paradox-Claim Audit, ADVERSARIALLY CORRECTED

Data cut: 1 September 2026. Plixorafenib / FORE8394 / PLX8394 is out of scope by instruction; no comparison is drawn or inferred anywhere below. Where a retrieved source contains a second-generation comparator, the comparator is redacted and not characterised.

Sources retrieved and read in full for this correction pass: ZELBORAF USPI (gene.com PDF, text-extracted locally), Zelboraf EMA SmPC Annex I (text-extracted locally), COTELLIC USPI (text-extracted locally), RCSB REST API for 3OG7, NCBI E-utilities and Europe PMC for all peer citations, and PMC full text for Poulikakos 2010, Yao 2015, Sievert 2013, Adelmann 2016 and Karoulia 2016.

1. Resolved identity

INN vemurafenib. Codes/aliases: PLX4032 (Plexxikon), RG7204 / RO5185426 (Roche), brand ZELBORAF. CAS 918504-65-1. Originator Plexxikon Inc.; co-developed and commercialised by Roche/Genentech — Genentech USA, Inc. is the US labeller; Roche Registration GmbH, Grenzach-Wyhlen, Germany is the EU MAH (SmPC §7, verified). EU marketing authorisation number EU/1/12/751/001.

PLX4720 is a different molecule — the tool/analogue compound that carries much of the foundational paradox literature. This distinction is load-bearing throughout and is the audit's single most important correct observation.

First FDA approval 17 Aug 2011 (NDA 202429), BRAF V600E-mutant unresectable/metastatic melanoma. [FLAG: approval date and NDA number were not re-verified against an FDA source in this pass.]
US indications now (USPI §1.1, §1.2, verified verbatim): melanoma with BRAF V600E; Erdheim-Chester Disease with BRAF V600 (note the broader V600 wording for ECD).
EU: date of first authorisation 17 February 2012; date of latest renewal 22 September 2016 (SmPC §9, verified).

Label currency. The gene.com USPI Highlights carries "Revised: 07/2026" — verified in the extracted text. A second stamp, "Revised: 11/2017", appears on the appended Medication Guide (patient labeling), which is separately dated; this is normal and is not a contradiction, contrary to how the audit framed its provenance caveat. The DailyMed SPL "Revised: 5/2020" discrepancy is real and remains unresolved. [FLAG: a human should reconcile gene.com 07/2026 against DailyMed before this PI is cited as "current".]

EU label currency cannot be fixed. SmPC §10 "DATE OF REVISION OF THE TEXT" prints no value — verified directly (the section contains only the EMA website pointer). Only first authorisation and latest renewal are establishable. Audit correct on this point.

2. Binding mode

ATP-competitive, mutant-BRAF-preferential inhibitor. Conventionally Type I½ (I.5): DFG-IN with the regulatory αC-helix displaced OUT.

This classification is now PRIMARY-GROUNDED, not review-grounded. The audit's honesty flag ("I did not read the Karoulia 2016 primary text… review-tier only") was a retrieval failure, not an absence. Karoulia Z, Wu Y, Ahmed TA, et al., Cancer Cell 2016;30(3):485–498, PMID 27523909, DOI 10.1016/j.ccell.2016.06.024, is free at PMC5021590 and was read for this correction. Verbatim from the primary:

"VEM uses a substituted sulphonamide group that extends inside the catalytic pocket stabilizing the αC-helix closer to the OUT position…"

"Thus, the position of the αC-helix stabilized in the drug-bound protomer results in reduced affinity of the drug for the second protomer due to a steric hindrance mechanism."

"The side-chain conformation of R506 is in a further OUT position in PB and VEM-bound structures, whereas αC-IN inhibitors stabilize R506 closer to IN position."

STRUCK — the R506 salt-bridge claim. The audit relayed "displaces R506 so as to disrupt a salt bridge stabilising the RAF dimer interface." Karoulia 2016 mentions no salt bridge in connection with R506. That formulation was introduced by the review-tier relay (Cope N, et al., Br J Cancer 2018, PMC5765234) and attributed to the primary. This is a review doing a peer source's job — the most damaging venue error in the audit. Removed.

Negative cooperativity is a structural inference, not a measurement. Karoulia supports reduced second-protomer affinity by co-crystal geometry ("due to a steric hindrance mechanism"). No direct binding or affinity determination for the second protomer was located. It must not be quoted as a measured cooperativity constant.

[FLAG: Karoulia 2016 carries an erratum — Cancer Cell 2016;30(3):501–503, DOI 10.1016/j.ccell.2016.08.008 — which the audit never checked and which was not assessed here. A human must read it before this mechanism is used.]

Co-crystal — CORRECTED. PDB 3OG7, "B-Raf Kinase V600E oncogenic mutant in complex with PLX4032", 2.45 Å, deposited 2010-08-16, released 2010-09-22. Primary citation Bollag G, et al., Nature 2010;467:596–599, PMID 20823850, DOI 10.1038/nature09454.

The asymmetric unit contains TWO chains (A and B), not one. RCSB REST API returns deposited_polymer_entity_instance_count = 2, assembly_count = 2. The audit's "asymmetric unit contains a monomer (C1, A1)" conflated assembly annotation with ASU content. Correct statement: two chains in the ASU resolving into two biological assemblies, each annotated monomeric (C1 / A1).
The deposited entity is an engineered construct. RCSB names the polymer entity "AKAP9-BRAF fusion protein" carrying at least 15 engineered solubilising mutations: I544A, I551K, Q562R, L588N, K630S, F667E, Y673S, A688R, L706S, Q709R, S713E, L716E, S720E. The residues the mechanism narrative depends on — L505, R506, R509 — are native and unmutated, so the contact claims survive; but the construct caveat must accompany any use of this structure.
No structure-to-behaviour inference is permitted here. Crystallographic assembly annotation says nothing about solution dimerisation biology. The audit placed the (incorrect) "monomer in the ASU" fact adjacent to its monomer-selectivity argument; that adjacency is removed.
PLX4720 has its own earlier co-crystal (3C4C), frequently and incorrectly cited for vemurafenib. Audit correct.

3. Monomer vs dimer

Vemurafenib is monomer-selective, not dimer-selective. It is the negative benchmark against which dimer-selectivity claims are made by others; it is a claimant in none.

LABEL-GROUNDED — EMA SmPC §5.1, Table 6 "Kinase inhibitory activity of vemurafenib against different BRAF kinases" (every value verified verbatim against the extracted Annex I):

Kinase	IC50 (nM)
BRAF V600E	10
BRAF V600K	7
BRAF V600R	9
BRAF V600D	7
BRAF V600G	8
BRAF V600M	7
BRAF V600A	14
BRAF wild-type	39

Only ~4-fold V600E-over-wild-type selectivity. Cellular anti-proliferation IC50 (SmPC §5.1, verbatim): V600-mutant lines 0.016 to 1.131 µM; BRAF wild-type lines 12.06 and 14.32 µM.

USPI §12.1 (verified verbatim): "Vemurafenib also inhibits other kinases in vitro such as CRAF, ARAF, wild-type BRAF, SRMS, ACK1, MAP4K5, and FGR at similar concentrations." The biochemical precondition for the paradox is disclosed in the label itself.

PEER-GROUNDED, monomer vs dimer:

Yao Z, et al., Cancer Cell 2015;28(3):370–383, PMID 26343582, PMC4894664, Figure 4A, verbatim: "more than 30-fold higher concentrations of vemurafenib were required to inhibit ERK driven by p61 V600E dimers than by p61 V600E R509H monomers." Cellular experiment in transfected SKBR3 cells. CORRECTED: the audit's parenthetical "monomer ~100–300 nM vs dimer >3,000 nM" is not in the source and is removed — an estimate the audit's own rule 4 forbids. Only the ratio is establishable.
Poulikakos PI, et al., Nature 2010;464:427–430, verbatim from PMC3178447: "PLX4032 inhibited…purified catalytic domains of BRAFV600E, wild-type BRAF and CRAF (IC50s: 35, 110 and 48nM)." CORRECTED framing: the audit reported these as bare biochemical IC50s while simultaneously asserting no purified-protein data exists. Precise position: purified catalytic-domain (monomer) enzymology exists; purified DIMER enzymology does not.

NOT DISCLOSED / NOT LOCATED. No biochemical potency measurement against purified, defined BRAF–CRAF heterodimers or BRAF homodimers. The >30-fold figure is a cellular comparison between an obligate-dimer p61 splice construct and its R509H dimerisation-deficient counterpart and must not be restated as a biochemical dimer selectivity ratio. No sponsor has published dimer-vs-monomer selectivity data for vemurafenib because no such selectivity is claimed.

4. Behaviour by alteration class

Monomeric BRAF V600E (class 1, low RAS-GTP) — INHIBITS. Label-grounded (SmPC §5.1 IC50 10 nM; USPI §12.1 "anti-tumor effects in cellular and animal models of melanomas with mutated BRAF V600E"). No paradoxical ERK activation reported in V600E cells with low RAS activity. Poulikakos 2010: HT-29 (endogenous V600E) is inhibited, but co-expressed mutant RAS confers resistance while MEK-inhibitor sensitivity is retained — the paradox is switched on by RAS activity, not by V600E status.

Dimeric BRAF fusions (KIAA1549-BRAF) — RESISTANT, WITH PARADOXICAL ACTIVATION — BUT NOT VEMURAFENIB EVIDENCE. Sievert AJ, et al., PNAS 2013;110:5957–5962, verbatim: "cell lines stably expressing KIAA1549-BRAF constructs were resistant to PLX4720 inhibition and/or displayed enhanced 'paradoxical' activation with increased phosphorylation of MEK and ERK1/2 in the presence of increasing concentrations of PLX4720." Signalling was abolished by dimer-interface mutagenesis; increased KSR binding was shown. The compound tested was PLX4720. Vemurafenib was never tested in this paper. No vemurafenib-specific pERK data in a BRAF-fusion model was located — inference by structural analogy only.

Class 2 (RAS-independent constitutive dimers — K601E, L597Q, G469A) — INSENSITIVE. Yao 2015 established that non-V600 activating BRAF mutants signal as constitutive RAS-independent dimers and are insensitive to vemurafenib, consistent with the >30-fold gap. Whether vemurafenib paradoxically activates ERK in class 2 cells, as distinct from merely failing to inhibit, was not established — unknown.

Class 3 (kinase-impaired/kinase-dead, RAS-dependent — D594G, G466V). Yao Z, et al., Nature 2017;548(7666):234–238, PMID 28783719, PMC5648058, DOI 10.1038/nature23291: class 3 mutants "bind more tightly than wild-type BRAF to RAS-GTP, and their binding to and activation of wild-type CRAF is enhanced, leading to increased ERK signalling"; they require coexistent RAS activation and are sensitive to inhibition of activated RAS. Vemurafenib-specific paradoxical pERK induction in class 3 models was not located — unknown. Mechanistically this is the highest-RAS-flux setting and therefore where paradox would be predicted; the prediction is not a finding and is not recorded as one.

RAS-mutant / BRAF wild-type contexts — PARADOXICAL ACTIVATION; best-evidenced category; vemurafenib itself tested.

Halaban R, et al., Pigment Cell Melanoma Res 2010;23(2):190–200, PMID 20149136, PMC2848976, DOI 10.1111/j.1755-148X.2010.00685.x — PLX4032 (vemurafenib) itself: activates ERK via RAF1 in BRAF wild-type melanoma cells; enhances proliferation of growth-factor-dependent NRAS-mutant primary melanoma cells; reduces adhesion, increases motility. The single most directly on-molecule preclinical paradox demonstration for vemurafenib in this record.
Yao 2015: pERK induction in SK-MEL-30 (wild-type RAF, NRAS-mutant). [FLAG: the audit cited "Figure S4G" for the low-concentration induction; the panel retrieved in this pass was Figure 4C (washout kinetics: "In SK-MEL-30, which contains WT RAF dimers, p-ERK rises rapidly, reaching a maximum approximately 1 hr after washout with a half-time of 10 minutes"). A human must confirm the exact panel before S4G is cited.]
Su F, et al., NEJM 2012: HRAS Q61L-mutant cell lines show increased proliferation on vemurafenib with MAPK signalling and ERK-mediated transcription.
Adelmann CH, et al., Oncotarget 2016: vemurafenib produced 6.86 ± 1.27-fold peak ERK induction in HaCaT-HRAS^G12V keratinocytes (verified verbatim).
NEW — T-cell context, beneficial direction, peer tier, vemurafenib itself, in vivo: Koya RC, Mok S, Otte N, et al., Cancer Res 2012;72(16):3928–3937, PMID 22693252, PMC3422880, DOI 10.1158/0008-5472.CAN-11-2837, verbatim: "vemurafenib paradoxically increased mitogen-activated protein kinase (MAPK) signaling, in vivo cytotoxic activity, and intratumoral cytokine secretion by adoptively transferred cells." The audit missed this entirely.
Clinical: progression of pre-existing NRAS-mutant CMML and KRAS-mutant pancreatic adenocarcinoma (both label-listed).

Foundational papers that used a tool compound rather than vemurafenib: Hatzivassiliou 2010 (GDC-0879, PLX4720), Heidorn 2010 (PLX4720, SB590885), Sievert 2013 (PLX4720), and Poulikakos 2010's definitive chemical-genetic panels (PLX4720 and the JAB34 orthogonal tool). Poulikakos did profile PLX4032 itself for potency and pERK induction. A substantial fraction of the "vemurafenib paradox" literature is in fact tool-compound literature, and the substitution is routinely elided downstream. This is the audit's strongest and best-supported observation and it is retained in full.

5. Clinical paradox signal

cuSCC / keratoacanthoma — USPI §5.1, Trial 1 (verified verbatim, Revised 07/2026): "The incidence of cutaneous squamous cell carcinomas (cuSCC) and keratoacanthomas in the ZELBORAF arm was 24% compared to < 1% in the dacarbazine arm." "The median time to the first appearance of cuSCC was 7 to 8 weeks; approximately 33% of patients who developed a cuSCC while receiving ZELBORAF experienced at least one additional occurrence with median time between occurrences of 6 weeks." Risk factors: age ≥65, prior skin cancer, chronic sun exposure. Denominators 336 vs 287. [FLAG: the USPI names no trial acronym — it says only "Trial 1". "BRIM-3" (EMA: NO25026) is an external mapping, not a label statement.]

cuSCC/KA in ECD — USPI §5.1, Trial 4 (verbatim): "the incidence of cuSCC and/or keratoacanthomas was 40.9% (9/22). The median time to first appearance of cuSCC amongst patients with at least one occurrence was 12.1 weeks."

New primary melanoma — USPI §5.1, Trial 1 (verbatim): "new primary malignant melanoma occurred in 2.1% (7/336) of patients receiving ZELBORAF compared to none of the patients receiving dacarbazine."

ADDED — myeloid neoplasms, USPI §5.1 (verbatim; omitted by the audit): "Cases of myeloid neoplasms amongst patients with ECD have been observed, including in patients who have received ZELBORAF. Monitoring complete blood count in ECD patients with co-existing myeloid malignancies is recommended."

Non-cutaneous SCC — USPI §5.1: warned about with no rate or denominator. PARTLY RESOLVED (audit recorded this as flatly undisclosed): the COTELLIC USPI §5.1 does give non-cutaneous malignancy rates — 1.2% in the vemurafenib monotherapy arm vs 0.8% in the combination arm (Trial 1).

EMA SmPC §4.8 (verified verbatim): "The incidence of cuSCC in vemurafenib-treated patients across studies was approximately 20%. The majority of the excised lesions reviewed by an independent central dermatopathology laboratory were classified as SCC-keratoacanthoma subtype or with mixed-keratoacanthoma features (52%)." ADDED — the companion figure the audit dropped: "Most lesions classified as 'other' (43%) were benign skin lesions (e.g. verruca vulgaris, actinic keratosis, benign keratosis, cyst/benign cyst)." Central review reclassified a large minority as benign; this materially qualifies any use of the raw incidence figures.

RAS-mutant tumour progression — label-listed. USPI §6.2 Postmarketing Experience, Neoplasms (verbatim): "Progression of pre-existing chronic myelomonocytic leukemia with NRAS mutation." CORRECTED / STRENGTHENED for the EU label: in the SmPC §4.8 tabulated adverse-reaction table, chronic myelomonocytic leukaemia and pancreatic adenocarcinoma both appear with an assigned frequency category of "Rare", with footnotes "(4) Progression of pre-existing chronic myelomonocytic leukaemia with NRAS mutation" and "(5) Progression of pre-existing pancreatic adenocarcinoma with KRAS mutation." These are frequency-categorised ADRs, not mere footnotes as the audit implied.

Index case with dechallenge/rechallenge — CITATION COMPLETED: Callahan MK, Rampal R, Harding JJ, et al., "Progression of RAS-mutant leukemia during RAF inhibitor treatment", N Engl J Med 2012;367(24):2316–2321, PMID 23134356, DOI 10.1056/NEJMoa1208958. NRAS-mutant CMML in a melanoma patient; leukocytosis and monocytosis after the first dose; temporal association across multiple dechallenge/rechallenge cycles; ex vivo vemurafenib exposure induced ERK hyperactivation and proliferation of the leukaemic population, reversed on withdrawal. n=1, but the rechallenge design plus the ex vivo mechanism make this the highest-causal-inference clinical evidence in the record.

Molecular confirmation in the lesions — Su F, et al., NEJM 2012;366(3):207–215, PMID 22256804, PMC3724537 (abstract verified verbatim): "Among 21 tumor samples, 13 had RAS mutations (12 in HRAS). In a validation set of 14 samples, 8 had RAS mutations (4 in HRAS). Thus, 60% (21 of 35) of the specimens harbored RAS mutations, the most prevalent being HRAS Q61L." CORRECTION to the audit's arithmetic gloss: the audit wrote "12 of 21 HRAS in the discovery set" — the source says 13 of 21 had RAS mutations, of which 12 were HRAS. Corroborated independently by Oberholzer PA, et al., J Clin Oncol 2012;30(3):316–321, PMID 22067401, DOI 10.1200/JCO.2011.36.7680. CONFLICT DISCLOSURE THE AUDIT OMITTED: Su 2012's first author affiliation is Hoffmann-La Roche, Nutley NJ; Plexxikon co-authors include Bollag, Zhang, Habets, Burton, Wong and Nguyen; funding is "Hoffmann-La Roche and others". Still peer tier — but the audit flagged Genentech's authorship of Hatzivassiliou while staying silent here.

"ACCELERATION NOT INITIATION" IS A DEMONSTRATED FINDING, NOT AN INTERPRETATION — the audit was wrong. Su 2012, verbatim: "In a mouse model of HRAS Q61L-mediated skin carcinogenesis, the vemurafenib analogue PLX4720 was not an initiator or a promoter of carcinogenesis but accelerated growth of the lesions harboring HRAS mutations, and this growth was blocked by concomitant treatment with a MEK inhibitor." This is an in vivo experiment addressing exactly the proposition the audit dismissed. Carry-over caveat: PLX4720, not vemurafenib.

MEK co-inhibition — COTELLIC USPI §5.1, Trial 1 (coBRIM), Revised 07/2026 (verified verbatim): "cutaneous squamous cell carcinoma (cuSCC) or keratoacanthoma (KA) (6% and 20%), basal cell carcinoma (4.5% and 2.4%), and second primary melanoma (0.8% and 2.4%)" for the COTELLIC-with-vemurafenib arm and the vemurafenib arm respectively. COUNTER-SIGNAL retained: basal cell carcinoma moved the opposite way. The label states incidences and asserts no mechanism. CORRECTED — the MEK-mitigation evidence is stronger than the audit allowed. The audit rested it solely on this label incidence and called the mechanistic reading a reader's interpolation. Su 2012 additionally shows MEK-inhibitor blockade of PLX4720-accelerated HRAS-mutant lesion growth in vivo. Two independent strands, not one.

Quantitative cross-agent correlation — Adelmann 2016 (every value verified verbatim against PMC5058692): paradox index defined as "the pERK activation EC80 divided by the IC80 against A375". Vemurafenib: index 5.5 (EC80 −5.58 ± 0.19 log[M] ≈ 2,100 nM; A375 IC80 −6.42 ± 0.24 log[M] ≈ 380 nM), clinical cuSCC 22%; dabrafenib index 10 / cuSCC 6%; encorafenib index 50 / cuSCC 3.7%. Assay cell line: HaCaT-HRAS^G12V. The authors claim a correlation ("correlates well with overall rates of cuSCC"), not a prediction, and caveat that their study "focused on RAS mutant cells" with "differences in in-vivo pharmacology and complementary mechanisms" beyond the assay. Rank-order concordance across three agents from separate trials with different populations and exposures — not a controlled comparison, not a validated predictor.

EVIDENCE AGAINST / LIMITS. No trial has isolated paradoxical MAPK activation as a prospective endpoint. The causal chain (drug → RAF dimer transactivation → ERK in HRAS-mutant keratinocytes → cuSCC) rests on (a) a randomised incidence difference, (b) RAS-mutation enrichment in lesions, (c) cell-based mechanism, (d) an in vivo mouse acceleration/MEK-blockade experiment using PLX4720, and (e) a clinical MEK-mitigation incidence signal. A strong assembled case — but still assembled, with no single confirmatory experiment in patients.

6. Paradox claims — corrected verdicts

#	Claim (abbrev.)	Who	Venue	Verdict	Change
1	"In vitro experiments have demonstrated paradoxical activation of MAP-kinase signaling and increased cell proliferation in BRAF wild-type cells…"	Genentech, USPI §5.2	label	supported	unchanged
2	"Based on mechanism of action, ZELBORAF may promote malignancies associated with activation of RAS…"	Genentech, USPI §5.1	label	supported	unchanged
3	"Based on mechanism of action, vemurafenib may cause progression of cancers associated with RAS mutations…"	Roche Registration GmbH, SmPC §4.4	label	supported	unchanged
4	cuSCC/KA 24% vs <1%; new primary melanoma 2.1% (7/336) vs 0; ECD 40.9% (9/22)	Genentech, USPI §5.1	label	supported	unchanged
5	MEK co-inhibition: cuSCC/KA 6% vs 20%; 2nd primary melanoma 0.8% vs 2.4%	Genentech, COTELLIC USPI §5.1	label	partially-supported	unchanged verdict; evidence base widened (Su 2012 in vivo MEK blockade)
6	ATP-competitive RAF inhibitors including PLX4032 transactivate RAF dimers; drug acts in trans on the unbound protomer; requires RAS, CRAF, intact dimer interface	Poulikakos 2010	peer	DOWNGRADED → partially-supported	verdict contradicted its own evidence_basis
7	PLX4032 activates ERK via RAF1 in BRAF-WT melanoma cells; enhances NRAS-mutant proliferation	Halaban 2010	peer	supported	unchanged
8	RAF inhibitors prime wild-type RAF to activate MAPK and enhance growth	Hatzivassiliou 2010 (Genentech)	peer	partially-supported	"independent replication" credit WITHDRAWN
9	BRAF inhibitors drive RAS-dependent BRAF–CRAF binding, CRAF activation, MEK-ERK signalling	Heidorn 2010	peer	partially-supported	"replication" credit WITHDRAWN
10	60% (21/35) of vemurafenib-associated cuSCC/KA carry RAS mutations, mostly HRAS Q61L	Su 2012	peer (sponsor-affiliated, Roche-funded)	supported	COI disclosed; arithmetic corrected; in vivo acceleration/MEK arm added
11	Vemurafenib caused accelerated progression of a pre-existing RAS-mutant malignancy; ERK hyperactivation reversed on withdrawal	Callahan 2012	peer	supported	citation completed
12	>30-fold higher vemurafenib concentrations needed against V600E dimers than monomers; pERK induced at low concentrations in WT-RAF/NRAS-mutant cells	Yao 2015	peer	supported	invented nM values struck
13	KIAA1549-BRAF homodimer resistant to PLX4720 with CRAF-independent paradoxical activation; paper contrasts a second-generation comparator against vemurafenib	Sievert 2013	peer	DOWNGRADED → asserted-only (for the vemurafenib-relevant content)	quotation not verbatim; vemurafenib never tested
14	Vemurafenib paradox index 5.5; tracks rank-order with clinical cuSCC across three agents	Adelmann 2016	peer	partially-supported	unchanged; all values verified
15	Vemurafenib stabilises αC-OUT; reduced drug affinity for the second protomer (negative cooperativity); ~~R506 salt-bridge disruption~~	Karoulia 2016	peer (venue corrected from "review")	UPGRADED → partially-supported	primary read; salt-bridge component STRUCK; mechanism is structural inference, not measurement
16	Canonical review framing of the RAF-inhibitor paradox	Holderfield 2014 NRC; Lito/Solit; Cope 2018	review	supported by the primaries it cites, not by itself	unchanged — awareness only
17	Vemurafenib-induced paradoxical MAPK activation has a beneficial immunological consequence (T-cell phenotype/function)	AACR 2012 Abstract 3510 (conference); Koya 2012 Cancer Res (peer)	conference + peer	UPGRADED → supported	audit's "unknown" was a retrieval failure; a peer paper testing vemurafenib itself in vivo exists
18	ABSENCE FINDING — no sponsor claim that vemurafenib avoids, breaks or mitigates paradoxical activation exists in the public record	Roche / Genentech / Plexxikon	company	"contradicted" → reworded to `no-claim-located`	a negative search result cannot carry a truth verdict
19	IMPLICIT LABEL CLAIM BY OMISSION — the MOA section describes vemurafenib purely as a kinase inhibitor, with no dimer/transactivation/paradox language	Genentech USPI §12.1; SmPC §5.1	label	asserted-only	partly qualified — see §1.1 Limitation of Use below

Detail on the downgrades and upgrades

#6 — Poulikakos 2010. Verified from PMC3178447: PLX4032 supplied purified catalytic-domain potency (35/110/48 nM) and appeared in the profiled panel; pERK induction ran 40 nM–1 µM with inhibition at 10 µM. But the definitive chemical-genetic experiments used PLX4720 and the JAB34 orthogonal tool, not PLX4032 — gatekeeper-mutant, kinase-dead V5-catC, and the R401 dimer-interface mutant all sit on tool compounds. So: "PLX4032 induces pERK" is supported; "PLX4032 acts in trans on the dimer partner" is inference by structural analogy. The audit's own evidence_basis said as much and then returned "supported" anyway — a claim/evidence conflation inside the audit itself.

#8 / #9 — the replication credit is withdrawn. Only Registered Reports exist. Bhargava A, et al., eLife 2016;5:e09976 (PMID 26882073, PMC4769155) for Hatzivassiliou — note the audit dated this 2015; it is 2016. And eLife 2016;5:e11999 (PMID 26885666, PMC4769162) for Heidorn. A Registered Report is a pre-registered study plan: "This Registered Report describes the proposed replication plan… the results of the replications will be published by eLife." No completed Replication Study was located for either paper; the replication outcome is unknown. The audit presented a protocol as a completed external check and called it "an unusually strong external check". [FLAG: a human should check the RP:CB aggregate reporting for whether either replication was ever completed or abandoned.]

#13 — Sievert 2013. The audit's quoted string "unlike vemurafenib, does not induce activation of wild-type BRAF" does not appear among the paper's three sentences containing "vemurafenib". The actual text is: "Whereas both vemurafenib and [comparator — out of scope, not characterised] display nearly identical, potent inhibition of phospho-ERK (pERK) in V600EBRAF-mutant cells, the well-described paradoxical activation associated with mutant RAS expression under vemurafenib treatment is absent upon treatment with [comparator]." Vemurafenib was not tested experimentally in this paper. The audit's own excellent observation stands and is reinforced: the paper treats vemurafenib's paradox as "the well-described paradoxical activation" — an unexamined cited premise, not a demonstrated result.

#17 — the beneficial-paradox claim is real and is peer-grounded. AACR Abstract 3510 content is publicly indexed: C-Raf implicated by co-immunoprecipitation and kinase assay in vemurafenib-induced paradoxical MAPK activation in T cells; C57BL6 mice dosed 3 weeks showed CD3+ skewing toward a central-memory-like phenotype (CD44+, CD62L+, LY-6C+). [FLAG: authors and affiliations for Abstract 3510 were still not established; the AACR site returned HTTP 403 on direct fetch and the record is absent from Europe PMC — treat the abstract itself as provisional.] The directional claim is independently carried at peer tier by Koya 2012 (above), which tested vemurafenib itself in vivo. Paradoxical activation is therefore not uniformly a liability in the published record — a counter-framing the audit registered only as an unverified rumour.

#19 — qualified. It remains true that "paradoxical", "dimer" and "transactivation" appear nowhere in USPI §12.1, that "dimer" appears nowhere in the USPI at all (grep-verified), and that "paradox" appears zero times in the entire EMA SmPC (grep-verified). But the audit's stronger assertion that the label "never explicitly joins" mechanism and consequence is too strong: USPI §1.1 Limitation of Use reads, verbatim, "ZELBORAF is not indicated for treatment of patients with wild-type BRAF melanoma [see Warnings and Precautions (5.2)]." The label ties the paradox warning directly to a prescribing restriction. The audit's point survives in its narrow form only: the label does not join §5.2 to the 24% cuSCC incidence in §5.1.

7. Unknowns — corrected

Genuine, verified unknowns (retained):

"Paradox" appears ZERO times in the entire EMA Zelboraf SmPC (grep-verified). The EU label states the clinical consequence (§4.4) without ever naming the mechanism; only the US label names it (§5.2). An asymmetry between two approved labels for the same molecule — and proof that "the label does not mention paradoxical activation" is a weak negative for any molecule. Retained; this is the audit's best finding.
"Dimer" appears nowhere in the ZELBORAF USPI (grep-verified). There is no label-level mechanistic account of the paradox for the canonical paradox-inducing agent.
USPI §5.2 cites no study for its "In vitro experiments have demonstrated…" statement. The underlying experiments are not identified in the label and were not traceable to a named dataset.
No purified-dimer biochemistry for vemurafenib exists. No IC50 against defined recombinant BRAF–CRAF heterodimers or BRAF homodimers was located. Purified catalytic-domain (monomer) values do exist (Poulikakos: 35/110/48 nM).
Class 2 paradox is unestablished — Yao 2015 establishes insensitivity, not paradoxical activation.
Class 3 paradox is unestablished for vemurafenib — Yao 2017 establishes the RAS-dependent, CRAF-transactivating biology; the vemurafenib-on-class-3 experiment was not found. Prediction is not evidence.
No vemurafenib-specific experiment in a BRAF-fusion model was located. Sievert used PLX4720.
Non-cuSCC of the head and neck has no rate in either label's warnings section. (Partly offset: COTELLIC §5.1 gives a 1.2% non-cutaneous malignancy rate for the vemurafenib arm.)
No Roche/Genentech press release, deck or corporate page was located making any independent mechanistic paradox claim beyond the label text hosted at gene.com. Whether non-indexed company material exists is unknown.
Bollag 2010 (Nature 467:596) was not read in full in either pass. Whether the sponsor's own discovery paper addressed paradoxical activation is unestablished. [FLAG]
DailyMed "Revised: 5/2020" vs gene.com "Revised: 07/2026" — unresolved. [FLAG]
EMA SmPC §10 prints no revision date — EU label currency at the 1 Sep 2026 cut cannot be fixed.
Causal attribution of the 24% cuSCC rate to paradoxical MAPK activation is nowhere demonstrated in a single confirmatory clinical experiment.
The Adelmann paradox index is an internally defined in vitro ratio with no regulatory or cross-lab validation.
The coBRIM basal-cell-carcinoma counter-signal (2.4% → 4.5%) is explained by neither label nor located literature.
AACR Abstract 3510 authors and affiliations remain unestablished. [FLAG]
Karoulia 2016's erratum (Cancer Cell 30:501–503) was not assessed. [FLAG]
Whether the RP:CB replications of Hatzivassiliou and Heidorn were ever completed is unknown — only protocols were located. [FLAG]
Plixorafenib / FORE8394 / PLX8394 is out of scope by instruction. Comparator data for that molecule in retrieved sources (Adelmann 2016's table; the Sievert 2013 comparator; the paradox-breaker literature) has been deliberately omitted, and no comparison has been drawn or inferred.

Removed from "unknown" because they were not unknown:

~~"Karoulia 2016 primary text was NOT read… therefore NOT primary-grounded"~~ — retrievable and read (PMC5021590).
~~"AACR 2012 Abstract 3510 returned HTTP 403; only the title is established"~~ — content publicly indexed, and the directional claim is peer-grounded by Koya 2012.
~~"The 'acceleration not initiation' reading is mechanistic interpretation, not a demonstrated finding"~~ — demonstrated in vivo in Su 2012 (with PLX4720).
~~"The MEK-inhibitor mitigation rests on a randomised incidence observation reported without mechanistic attribution"~~ — an in vivo MEK-blockade experiment exists (Su 2012).

8. Citations — corrected

LABEL - ZELBORAF (vemurafenib) US Prescribing Information, Genentech, Inc., Revised: 07/2026 (Highlights stamp; appended Medication Guide separately dated 11/2017). §1.1 Limitation of Use; §5.1 New Primary Malignancies; §5.2 Tumor Promotion in BRAF Wild-Type Melanoma; §6.2 Postmarketing Experience; §12.1 Mechanism of Action. https://www.gene.com/download/pdf/zelboraf_prescribing.pdf (NDA 202429). Retrieved and text-extracted for this review. - Zelboraf 240 mg film-coated tablets, EMA SmPC (EPAR product information, Annex I), Roche Registration GmbH, EU/1/12/751/001. §4.4; §4.8 (tabulated ADRs incl. CMML and pancreatic adenocarcinoma at frequency "Rare", footnotes 4 and 5; cuSCC narrative incl. 52% KA-subtype and 43% "other"/benign); §5.1 Table 6; §9 (first authorisation 17 Feb 2012, latest renewal 22 Sep 2016); §10 prints no revision date. https://www.ema.europa.eu/en/documents/product-information/zelboraf-epar-product-information_en.pdf Retrieved and text-extracted for this review. - COTELLIC (cobimetinib) US Prescribing Information, Genentech, Inc., Revised: 07/2026. §5.1 New Primary Malignancies — Cutaneous Malignancies (cuSCC/KA 6% and 20%; BCC 4.5% and 2.4%; second primary melanoma 0.8% and 2.4%) and Non-Cutaneous Malignancies (0.8% and 1.2%). https://www.gene.com/download/pdf/cotellic_prescribing.pdf Retrieved and text-extracted for this review.

PEER - Poulikakos PI, Zhang C, Bollag G, Shokat KM, Rosen N. Nature 2010;464(7287):427–430. DOI 10.1038/nature08902. PMID 20179705. PMC3178447. (Bollag = Plexxikon.) - Hatzivassiliou G, Song K, Yen I, et al. Nature 2010;464(7287):431–435. DOI 10.1038/nature08833. PMID 20130576. (Genentech; compounds GDC-0879 and PLX4720, not vemurafenib.) - Heidorn SJ, Milagre C, Whittaker S, et al. Cell 2010;140(2):209–221. DOI 10.1016/j.cell.2009.12.040. PMID 20141835. PMC2872605. (PLX4720 and SB590885.) - Halaban R, Zhang W, Bacchiocchi A, et al. Pigment Cell Melanoma Res 2010;23(2):190–200. DOI 10.1111/j.1755-148X.2010.00685.x. PMID 20149136. PMC2848976. (Vemurafenib itself tested.) [FLAG: PubMed renders the title as "…proliferation of BRAF melanoma cells"; the published title carries a superscript "WT" that indexing strips. Cosmetic.] - Su F, Viros A, Milagre C, et al. N Engl J Med 2012;366(3):207–215. DOI 10.1056/NEJMoa1105358. PMID 22256804. PMC3724537. First-author affiliation Hoffmann-La Roche; Plexxikon co-authors; funded by Hoffmann-La Roche. - Oberholzer PA, Kee D, Dziunycz P, et al. J Clin Oncol 2012;30(3):316–321. PMID 22067401. DOI 10.1200/JCO.2011.36.7680. - Callahan MK, Rampal R, Harding JJ, et al. N Engl J Med 2012;367(24):2316–2321. DOI 10.1056/NEJMoa1208958. PMID 23134356. PMC3627494. - Yao Z, Torres NM, Tao A, et al. Cancer Cell 2015;28(3):370–383. DOI 10.1016/j.ccell.2015.08.001. PMID 26343582. PMC4894664. - Yao Z, Yaeger R, Rodrik-Outmezguine VS, et al. Nature 2017;548(7666):234–238. DOI 10.1038/nature23291. PMID 28783719. PMC5648058. - Sievert AJ, Lang S-S, Boucher KL, et al. PNAS 2013;110(15):5957–5962. DOI 10.1073/pnas.1219232110. PMID 23533272. PMC3625308. (PLX4720; vemurafenib not tested.) - Adelmann CH, Ching G, Du L, et al. Oncotarget 2016;7(21):30453–30460. DOI 10.18632/oncotarget.8351. PMID 27028853. PMC5058692. - Karoulia Z, Wu Y, Ahmed TA, et al. Cancer Cell 2016;30(3):485–498. DOI 10.1016/j.ccell.2016.06.024. PMID 27523909. PMC5021590. — MOVED FROM "review, not read" TO PEER, READ IN FULL. Erratum: Cancer Cell 2016;30(3):501–503, DOI 10.1016/j.ccell.2016.08.008 [not assessed]. (C. Zhang and G. Bollag are Plexxikon employees, declared.) - Koya RC, Mok S, Otte N, et al. "BRAF inhibitor vemurafenib improves the antitumor activity of adoptive cell immunotherapy." Cancer Res 2012;72(16):3928–3937. DOI 10.1158/0008-5472.CAN-11-2837. PMID 22693252. PMC3422880. — NEW; MISSED BY THE AUDIT. - Bollag G, Hirth P, Tsai J, et al. Nature 2010;467:596–599. DOI 10.1038/nature09454. PMID 20823850. (Primary citation for PDB 3OG7. NOT read in full in either pass.)

PROTOCOL — NOT a replication result - Bhargava A, Pelech S, Woodard B, Kerwin J, Maherali N; Reproducibility Project: Cancer Biology. "Registered report: RAF inhibitors prime wild-type RAF to activate the MAPK pathway and enhance growth." eLife 2016;5:e09976. DOI 10.7554/eLife.09976. PMID 26882073. PMC4769155. - Bhargava A, Anant M, Mack H; Reproducibility Project: Cancer Biology. "Registered report: Kinase-dead BRAF and oncogenic RAS cooperate to drive tumor progression through CRAF." eLife 2016;5:e11999. DOI 10.7554/eLife.11999. PMID 26885666. PMC4769162.

STRUCTURE - RCSB PDB 3OG7, "B-Raf Kinase V600E oncogenic mutant in complex with PLX4032". 2.45 Å; deposited 2010-08-16; released 2010-09-22. Two polymer chains (A, B) in the asymmetric unit; two biological assemblies, each annotated monomeric (C1/A1). Polymer entity described by RCSB as "AKAP9-BRAF fusion protein" with ≥15 engineered mutations (I544A, I551K, Q562R, L588N, K630S, F667E, Y673S, A688R, L706S, Q709R, S713E, L716E, S720E). The entry does not annotate αC-helix conformation. https://www.rcsb.org/structure/3OG7 (verified via data.rcsb.org REST API.)

REVIEW — AWARENESS ONLY, cannot establish or refute a mechanism - Holderfield M, Deuker MM, McCormick F, McMahon M. Nature Reviews Cancer 2014 (nrc3760). - Cope N, et al. "Targeting the Raf kinases in human cancer: the Raf dimer dilemma." Br J Cancer 2018. PMC5765234. ⚠ This review is the origin of the R506 "salt bridge" formulation that is NOT in Karoulia 2016. Do not cite it as a proxy for the primary. - "The RAF inhibitor paradox revisited." PMID 22340588. PMC3285264.

CONFERENCE — provisional - Abstract 3510: "Paradoxical MAPK activation and beneficial effects of vemurafenib on T-cell phenotype resulting in improved functionality in vivo." AACR Annual Meeting 2012 (103rd, Chicago); Cancer Res 2012;72(8_Supplement):3510. Content partially recovered via public indexing; authors and affiliations not established; direct fetch returns HTTP 403. [FLAG]

9. Bottom line for downstream use

Vemurafenib remains the reference paradox-inducing agent and the negative benchmark for dimer-selectivity claims. That conclusion survives adversarial review intact — indeed the label evidence is stronger than the audit recorded (frequency-categorised RAS-malignancy ADRs in the EU; an indication-level restriction cross-referenced to the paradox warning in the US; an in vivo acceleration-and-MEK-blockade experiment in the peer record).

But four things in the audit must not be carried forward as written:

A review was allowed to speak for a primary. The R506 salt-bridge mechanism is a review artefact. Strike it.
A study protocol was reported as a completed replication. Withdraw the "independently replicated" credit from Hatzivassiliou and Heidorn.
Two "unknowns" were retrieval failures, not absences — Karoulia 2016 and the beneficial-paradox T-cell claim. One of them reverses a verdict from "unknown" to "supported" on peer evidence testing vemurafenib itself in vivo.
A quotation does not match its source (Sievert), and a set of nanomolar values was invented (Yao 2015).

The audit's most valuable contribution — that a large share of the "vemurafenib paradox" literature is PLX4720 tool-compound literature, and that this substitution is elided downstream — is correct, verified, and strengthened by this pass. Poulikakos 2010's definitive chemical-genetic panels join Hatzivassiliou, Heidorn and Sievert on the tool-compound side of that line.

AUnknown register 98

What could not be established from public sources. These are findings, not omissions.

VEMURAFENIB — purified-dimer biochemistry: NOT DISCLOSED. No IC50 against defined recombinant BRAF-CRAF heterodimers or BRAF homodimers was located. Purified catalytic-domain (monomer) values do exist (35 / 110 / 48 nM).
VEMURAFENIB — class 2 BRAF (K601E, L597Q, G469A) paradoxical activation: UNKNOWN. Yao 2015 establishes insensitivity, not paradoxical activation.
VEMURAFENIB — class 3 BRAF (D594G, G466V) paradoxical activation: UNKNOWN. The vemurafenib-on-class-3 experiment was not found. Mechanistically this is the highest-RAS-flux setting and therefore where paradox would be predicted; the prediction is not a finding and is not recorded as one.
VEMURAFENIB — BRAF fusion (KIAA1549-BRAF) behaviour: UNKNOWN for vemurafenib. Sievert 2013 used PLX4720; vemurafenib was never tested in that paper.
VEMURAFENIB — the studies underlying USPI 5.2 ('In vitro experiments have demonstrated...'): NOT IDENTIFIED. The label cites no study and the experiments were not traceable to a named dataset.
VEMURAFENIB — mechanistic account of the paradox at label level: ABSENT. 'Dimer' appears nowhere in the ZELBORAF USPI; 'paradox' appears ZERO times in the entire EMA SmPC.
VEMURAFENIB — non-cutaneous SCC of the head and neck: no rate or denominator in either label's warnings section. (Partly offset: COTELLIC 5.1 gives 1.2% non-cutaneous malignancy for the vemurafenib arm.)
VEMURAFENIB — causal attribution of the 24% cuSCC rate to paradoxical MAPK activation: NOWHERE DEMONSTRATED in a single confirmatory clinical experiment.
VEMURAFENIB — coBRIM basal-cell-carcinoma counter-signal (2.4% to 4.5% on adding cobimetinib): EXPLAINED BY NEITHER LABEL NOR LOCATED LITERATURE.
VEMURAFENIB — whether the RP:CB replications of Hatzivassiliou 2010 and Heidorn 2010 were ever completed: UNKNOWN. Only protocols (Registered Reports) were located.
VEMURAFENIB — EU label currency at the 1 Sep 2026 cut: CANNOT BE FIXED. SmPC section 10 'DATE OF REVISION OF THE TEXT' prints no value.
VEMURAFENIB — DailyMed 'Revised: 5/2020' vs gene.com 'Revised: 07/2026': UNRESOLVED.
VEMURAFENIB — Bollag 2010 (Nature 467:596, the sponsor's own discovery paper and primary citation for PDB 3OG7): NOT READ IN FULL. Whether it addressed paradoxical activation is unestablished.
VEMURAFENIB — Karoulia 2016 erratum (Cancer Cell 30:501-503): NOT ASSESSED.
VEMURAFENIB — AACR 2012 Abstract 3510 authors and affiliations: UNESTABLISHED (AACR returns HTTP 403).
VEMURAFENIB — any Roche/Genentech press release, deck or corporate page making an independent mechanistic paradox claim beyond the label text: NONE LOCATED. Whether non-indexed company material exists is unknown.
DABRAFENIB — sponsor-disclosed dimer IC50 or monomer:dimer selectivity ratio: NEVER PUBLISHED. The only quantitative figure (30-100x) is from an independent academic lab and applies to PB, DAB and VEM alike in that experiment.
DABRAFENIB — activity against KIAA1549::BRAF or any dimeric BRAF fusion: NO PUBLICATION TESTS IT. Confirmed across Sievert 2013 and Jain 2017, neither of which tested dabrafenib. Genuinely unmeasured.
DABRAFENIB — prospective monotherapy dataset in class II or class III BRAF mutants: NONE EXISTS, and no NCI-MATCH arm can supply one (EAY131-H is V600-restricted; sub-protocol R is trametinib-only). Every class II clinical datapoint located is dabrafenib PLUS trametinib.
DABRAFENIB — residue-level contact map beyond R506: NOT ESTABLISHED. The four PDB entries are verified but their primary papers were not read.
DABRAFENIB — denominators behind the pooled US monotherapy percentages (cuSCC 11%, KA 4%, BCC 4%, new primary melanoma 1%): NOT EXTRACTED. USPI 6.1 remains unread — a retrievable gap in the same PDF, not an absence in the record.
DABRAFENIB — RAS-genotype confirmation for the 1% non-cutaneous malignancies: NOT STATED in either label, in either jurisdiction.
DABRAFENIB — BREAK-3 (Hauschild 2012) and Falchook 2012 cuSCC rates: UNVERIFIED. Secondary sources conflict (6% vs 10%); a company release gives 7%. NEJM's 9% and the SmPC's 10% (22/211) for the same COMBI-d arm do not agree.
DABRAFENIB — conference-tier material making a paradox claim beyond what was subsequently published peer-reviewed: NONE LOCATED (recorded as 'none located', not 'none exists').
DABRAFENIB — head-to-head clinical trial against any other BRAF inhibitor as monotherapy: NONE EXISTS. Every cross-agent cuSCC comparison in the record is cross-trial and unadjusted for surveillance intensity.
ENCORAFENIB — co-crystal structure with BRAF: NONE EXISTS (verified absence). Every binding-mode statement is a docked pose on the BRAF-dabrafenib template 5CSW. No structural evidence exists of how encorafenib occupies a RAF dimer.
ENCORAFENIB — quantitative monomer-versus-dimer potency: NOT DISCLOSED. No side-by-side IC50, no dimer-selectivity ratio, no measured cooperativity coefficient. Verified absent across both candidate sources (Clayton 2025 names it once with no values; Tkacik 2026 did not test it).
ENCORAFENIB — activity against BRAF FUSION proteins: UNKNOWN. No data in any dimer-constitutive fusion model in any venue.
ENCORAFENIB — link between residence time and PARADOX behaviour: NO PRIMARY EXPERIMENT. The >30 h figure is measured from V600E-MUTANT BRAF, not from wild-type BRAF or CRAF where paradox occurs. No study varies residence time and measures paradoxical pERK induction as the dependent variable.
ENCORAFENIB — Stuart AACR 2012 abstract 3790 full text: NON-RETRIEVABLE (HTTP 403). Four of six facts previously sourced to it (>400-cell-line screen, >30 h off-rate, gastric-hyperplasia surrogate, A375 pERK/antiproliferative EC50s) are not in the retrievable text and reach the record only through review paraphrase. Comparator off-rates (2 h dabrafenib, 0.5 h vemurafenib) share this dependency.
ENCORAFENIB — trial-level or series-level rate of RAS-mutant tumour progression on drug: NONE PUBLISHED, despite both labels mandating surveillance and discontinuation. The regulators require the behaviour without disclosing an incidence.
ENCORAFENIB — rationale for the EU label's silence on paradox: UNKNOWN from public documents. No EMA assessment-report rationale for the omission was located.
ENCORAFENIB — whether the paradox differs QUANTITATIVELY from dabrafenib in a controlled setting: ONLY WEAKLY ESTABLISHED. No randomised head-to-head trial reporting cuSCC as an endpoint exists.
ENCORAFENIB — COLUMBUS vemurafenib-arm cuSCC/KA rate: NOT RETRIEVED HERE (Lancet Oncol and the JCO 5-year update both returned 403). This is the cleanest available randomised clinical paradox contrast and it is probably obtainable.
ENCORAFENIB — denominators for every FDA 5.1 cutaneous-malignancy row: NOT STATED IN THE LABEL. Candidate denominators exist only as a derived table.
MOSPERAFENIB — binding mode: ENTIRELY UNDISCLOSED. No type assignment, no DFG/alphaC/R506 status, no contact residues, no co-crystal, no PDB ID.
MOSPERAFENIB — monomer-versus-dimer potency: NEVER PUBLISHED IN READABLE FORM. No dimer-disruption assay, forced-dimer comparison or selectivity ratio is publicly readable. A review's dimer column headed 'confirmed experimentally' enters BRAF/BRAF and cites two paywalled 2022 papers, so such data may exist behind the paywall.
MOSPERAFENIB — the foundational paradox experiment (Wichmann 2022, Clin Cancer Res 28:770-780): NOT PUBLICLY VERIFIABLE. Full text is a Cloudflare challenge page; the abstract names no cell line, no concentration and no assay. It is simultaneously the origin of the paradox designation, the cited source of the vendor affinity figures, and the review's cited source for the BRAF/BRAF dimer entry.
MOSPERAFENIB — cSCC status in the final n=80 dataset: UNRESOLVED. Two conference abstracts assert absence; the peer-reviewed paper is silent on cSCC. JCO full text Cloudflare-blocked.
MOSPERAFENIB — new primary melanoma rate: NEVER REPORTED at any data cut, in any venue.
MOSPERAFENIB — human pharmacodynamic evidence of the mechanism: NONE OF ANY KIND. All clinical pERK figures are PK-derived. No skin biopsy, no paired tumour biopsy, no normal-tissue pERK anywhere.
MOSPERAFENIB — clinical data in RAS-mutant disease: NONE, AND NONE POSSIBLE FROM THIS TRIAL. Eligibility was BRAF V600-mutant only; no RAS-mutant patient was ever dosed.
MOSPERAFENIB — data in dimeric BRAF fusions, p61 splice variants, class II or class III BRAF: NONE.
MOSPERAFENIB — NRAS- or HRAS-mutant model, melanocyte, keratinocyte or skin model: NONE. The published wild-type panel is three colorectal lines.
MOSPERAFENIB — the patent family member that actually claims the molecule: NOT IDENTIFIED. WO2021116050A1 expressly disclaims it; an INPADOC family search on the excluded structure or CAS 2649372-20-1 is the next step, and that document would carry any per-compound paradox assay values.
MOSPERAFENIB — EMA absence: NOT ESTABLISHED. Record as 'not retrieved', not as 'searched and absent'.
MOSPERAFENIB — the JCO letter/reply exchange (the only independent peer-reviewed challenge on record): NOT READ. Neither item has an abstract in Europe PMC; both Cloudflare-blocked. It cannot be excluded that safety or paradox is discussed.
MOSPERAFENIB — ESMO-family abstract 126P: NOT READ (HTTP 403).
MOSPERAFENIB — reason for removal from Roche's phase I pipeline: NOT DISCLOSED. Roche states the fact of removal and gives no reason; portfolio prioritisation, competitive positioning and strategic reasons are all equally consistent with the disclosure as written.
MOSPERAFENIB — independent (non-Roche) replication: NONE EXISTS. Every paradoxical-activation experiment on this molecule in the public record was performed or co-authored by Roche.
NXP200 — binding mode: ENTIRELY UNKNOWN. No type I / I-and-a-half / II designation, no DFG or alphaC configuration, no contact residues, no co-crystal, no PDB ID, no SPR/ITC data.
NXP200 — chemical structure: NOT DISCLOSED (verified: PubChem 'No CID found', ChEMBL 0 records, no attributable Haisco patent).
NXP200 — monomer-vs-dimer potency: NOT DISCLOSED. No paired IC50s, no CRAF or heterodimer values, no biochemical dimer-affinity measurement. Sole potency figure in the entire record: BRAF V600E enzymatic IC50 = 5 nM.
NXP200 — fold-selectivity over wild-type BRAF: ASSERTED BUT NEVER QUANTIFIED (and asserted at the cellular, not enzymatic, level).
NXP200 — magnitude of 'significantly less / reduced paradoxical activation': NOWHERE QUANTIFIED. The sponsor's central peer-facing claim has no number attached to it in any source.
NXP200 — the dimerisation assay: NOT NAMED. No method, dose-response, quantification or blot is public in AACR 2024 #598; LB405's 'disrupts dimers' statement likewise carries no method or data.
NXP200 — paradox testing in any second RAS-mutant line, in BRAF-wild-type keratinocytes/skin models, or in vivo: NONE. HCT116 remains a single-line, single-comparator observation after full re-retrieval.
NXP200 — class III mutants and BRAF fusions: ZERO DATA, preclinical or clinical. Class II is asserted at conference tier with no mutant named and no data shown.
NXP200 — clinical pharmacodynamic data (tumour or skin pERK) at any dose in any study: NONE, despite PD being a stated secondary objective of NCT06536400. (Preclinical tumour pERK IHC does exist.)
NXP200 — cuSCC, keratoacanthoma, hyperkeratosis, skin papilloma and new primary malignancy: NEITHER REPORTED NOR EXCLUDED in any of the three clinical abstracts. Median follow-up, exposure duration and dermatological surveillance methodology all undisclosed.
NXP200 — RAS-mutant tumour-progression signal: NOT OBSERVABLE BY DESIGN. All four registered trials require BRAF V600.
NXP200 — posters and slide decks for all four abstracts: NOT IN HAND. Every 'method not named' and 'no data shown' finding is a statement about the abstract text only.
NXP200 — NSCLC and papillary thyroid 'durable responses': NO SUPPORTING NUMBERS, DENOMINATORS OR DURATIONS ANYWHERE.
NXP200 — Nuvectis 8-K exhibit 99.1 (accession 0001104659-26-090094): HTTP 403; corporate-deck paradox language unaudited. The defined licence 'Territory' is likewise unconfirmed.
NXP200 — ESMO 2026 (Madrid, 23-27 Oct 2026) oral presentation of additional Haisco clinical data: NOT PUBLIC at this cut; the most likely near-term source of paradox-relevant safety data.
NXP200 — INN: NONE ASSIGNED. No US IND filed (expected 4Q2026). No regulatory review document of any kind exists to audit in any jurisdiction.
NXP200 — independent laboratory characterisation: NONE. All preclinical and all paradox-specific data originate from Haisco employees.
CLATURAFENIB — label tier: EMPTY, and none can exist until approval (verified by retrieval, not assumed).
CLATURAFENIB — any public source calling it a 'paradox breaker': NONE. The class label is an internal Fore attribution unmirrored in the public record.
CLATURAFENIB — magnitude of residual paradoxical induction: NOT REPORTED. No fold-change, induction EC50 or paradox index in the text; Supplementary Fig. S1A not inspected.
CLATURAFENIB — paradox index: NEVER RUN. The molecule cannot be placed on the standard comparative scale.
CLATURAFENIB — RAS-mutant / BRAF-wild-type paradox experiment: NONE EXISTS. No such cell line, no xenograft, no in vivo model, no skin/keratinocyte or 3D skin-equivalent, no patient cohort (absent by inclusion criterion, NOT by an explicit RAS exclusion). The single largest evidentiary gap: the defining context for paradoxical activation has not been tested.
CLATURAFENIB — BRAF fusion data: NONE preclinically; fusion enrolment among the 73 unreported patients in NCT05355701 is UNKNOWN, not confirmed absent.
CLATURAFENIB — monomer-vs-dimer biochemical potency comparison: NONE. No enforced-dimer assay, no selectivity ratio, no SPR/SEC-MALS/AUC/cross-linking. The mechanism rests on a hedged structural inference plus non-matched co-IP.
CLATURAFENIB — endogenous wild-type dimer experiment: NOT REPORTED. The wild-type dimer co-IP is an over-expression experiment in HEK293T.
CLATURAFENIB — binding-mode type: NOT ASSIGNED. DFG state never stated; contact residues not disclosed.
CLATURAFENIB — PDB 9BP8 linkage to the primary paper: CORROBORATED BUT NOT FORMAL. Shared depositor/author and matching construct length; primary citation still 'To be published'; no accession in Data Availability. No dedicated medicinal-chemistry or structural discovery paper exists.
CLATURAFENIB — cutaneous-neoplasia rate at any denominator: NONE. Monotherapy denominator 18; AE threshold 3 or more patients; no surveillance protocol; NCT05355701 unposted.
CLATURAFENIB — exposure duration: NOT REPORTED. Latency to paradox-driven cutaneous neoplasia cannot be assessed.
CLATURAFENIB — class II/III clinical activity: UNPROVEN. Zero responses in eight patients (4 class II, 4 class III); only one at the efficacious dose.
CLATURAFENIB — patent literature: NEVER SEARCHED. Array/Pfizer composition-of-matter and method-of-use filings routinely contain BRAF-wild-type pERK and paradox data that never reach a journal. An entire unexamined evidence stream.
CLATURAFENIB — independent (non-sponsor) experimental replication: NONE EXISTS. Every preclinical, structural, target-engagement and co-IP experiment is by Pfizer Boulder R&D authors. The compound is sold by at least eight commercial vendors, so independent profiling is feasible — its absence is a genuine gap, not an access constraint.
TOVORAFENIB — cell-based paradox index: DOES NOT EXIST, from any source. Verified absent from Adelmann 2016. No equivalent HRAS-mutant keratinocyte assay published by anyone.
TOVORAFENIB — head-to-head CELLULAR monomer-vs-dimer potency comparison: NOT DISCLOSED by anyone. The EU SmPC 5.1 'monomers and dimers' phrase rests on crystallography and single-isoform enzymatic IC50s.
TOVORAFENIB — class II BRAF point mutants: NO POTENCY OR pERK DATASET. Only two BRAF fusion models were submitted to EMA.
TOVORAFENIB — class III BRAF: COMPLETE EVIDENTIARY VOID. Highest theoretical paradox risk; nothing published.
TOVORAFENIB — prospective clinical study in a RAS-mutant population with paired skin or tumour pERK biopsies: NONE. The only RAS-mutant exposure is inside the phase 1, with tumour pERK but no skin pharmacodynamics.
TOVORAFENIB — three-year FIREFLY-1 cuSCC / KA / SPM denominator: NOT DISCLOSED. Only the 10 May 2024 figures are public.
TOVORAFENIB — LOGGIC/FIREFLY-2 safety data that would test cutaneous malignancy over longer treatment-naive exposure: NONE PUBLISHED. EMA has deferred FIREFLY-2 results until July 2030.
TOVORAFENIB — FAERS disproportionality analysis, PV signal assessment or case report on tovorafenib and cuSCC/KA/new primary melanoma: NONE as of 1 September 2026. Absence of a published analysis is not evidence of absence of a signal.
TOVORAFENIB — whether the low-concentration pERK rise occurs at drug levels reached in HUMAN SKIN: NOT ESTABLISHED. Neither label nor sponsor paper maps 'clinically relevant concentrations' to measured human skin or plasma unbound concentrations on the approved schedule.
TOVORAFENIB — magnitude and fold-change of the NF1-LOF pERK rise: NOT QUANTIFIED in the accessible text; described qualitatively, and CHMP notes the trend differed between the NF1-LOF cell lines.
TOVORAFENIB — ARAF potency: UNRECONCILED BY MORE THAN 50-FOLD. Sponsor 55 nM vs Tkacik >3000 nM. ARAF was not submitted to EMA at all. Since ARAF sparing is the mechanism both the sponsor and CHMP invoke to explain the NF1-LOF pERK rise, this is a material gap in the only proposed mechanism.
TOVORAFENIB — independent replication of the founding HRAS-G12V paradox experiments (Sun 2017): NOT PUBLISHED.
TOVORAFENIB — the FDA regulatory record: NEVER RETRIEVED. Until the NDA 217700 Multi-Discipline Review and the FDA Approval Summary (Clin Cancer Res 2025;31(8):1383-1389) are read, 'the US regulator made no paradox statement' is silence, not verified absence. Only the US LABEL has been verified.
CROSS-MOLECULE — no randomised, prospectively dermatologically-surveilled head-to-head trial of any claimed paradox breaker against any early-generation BRAF inhibitor exists in any jurisdiction. Every cross-agent cutaneous comparison in this entire audit is cross-trial and unadjusted.
CROSS-MOLECULE — no molecule other than vemurafenib, dabrafenib and encorafenib has ever been run through a published paradox-index assay, so none of the claimed paradox breakers can be placed on the only comparative in vitro scale that exists.
CROSS-MOLECULE — no paradox-breaker claimant has any published human tissue pharmacodynamic data (skin or normal-tissue pERK) of any kind.

BOpen flags 74

[FLAG: a human should reconcile gene.com 07/2026 against DailyMed before this PI is cited as "current".]
[FLAG: Karoulia 2016 carries an erratum — Cancer Cell 2016;30(3):501–503, DOI 10.1016/j.ccell.2016.08.008 — which the audit never checked and which was not assessed here. A human must read it before this mechanism is used.]
[FLAG: the audit cited "Figure S4G" for the low-concentration induction; the panel retrieved in this pass was Figure 4C (washout kinetics: "In SK-MEL-30, which contains WT RAF dimers, p-ERK rises rapidly, reaching a maximum approximately 1 hr after washout with a half-time of 10 minutes"). A human must confirm the exact panel before S4G is cited.]
[FLAG: the USPI names no trial acronym — it says only "Trial 1". "BRIM-3" (EMA: NO25026) is an external mapping, not a label statement.]
[FLAG: a human should check the RP:CB aggregate reporting for whether either replication was ever completed or abandoned.]
[FLAG: authors and affiliations for Abstract 3510 were still not established; the AACR site returned HTTP 403 on direct fetch and the record is absent from Europe PMC — treat the abstract itself as provisional.]
[FLAG: human must confirm the audit's original Sievert string does not appear in a figure legend or supplement.]
[FLAG: PubMed renders the Halaban title as "…proliferation of BRAF melanoma cells"; the published title carries a superscript "WT" that indexing strips. Cosmetic.]
[FLAG: if a binding-mode class is needed for a comparison table, state which review's taxonomy is being used and cite it as review tier.]
[FLAG: any comparison table built on "dabrafenib loses 30–100× on dimers" must not imply the comparators do not.]
[FLAG: monomer IC50 "~100–300 nM" in SKBR3 not verified this pass — do not quote.]
[FLAG: re-read Yao 2017 at page level before quoting; and note this study is class-level — dabrafenib is not separately resolved, so it carries the same class-inference limitation the audit correctly applies to fusions.]
[FLAG: "supported" for TAFINLAR §5.2 means the label makes this claim, verified. It is not a label-tier experimental demonstration for dabrafenib specifically.]
[FLAG: how many of the 1% non-cutaneous malignancies were RAS-confirmed is not stated in either label.]
[FLAG: do not report the SmPC §4.4 named tumour types as genotype-confirmed RAS-mutant tumours.]
[FLAG: state which source is being quoted for COMBI-d cuSCC; do not present the SmPC figure (10%, 22/211) as "corroborating" the NEJM figure (9%).]
[FLAG: priority/dimerisation potency and clinical cuSCC incidence are different axes — do not present Karoulia's priming result as a direct rebuttal of a toxicity-incidence claim.]
[FLAG: the audit built its critique of Holderfield on a quote it had altered — the review had already marked this as speculation.]
[FLAG: the dabrafenib negative ("no dimer-selectivity or paradox-breaking claim exists") is usable as "none located in this pass", never as "none exists".]
[FLAG: the 7% / 2% (3/187) BREAK-3 figures are company tier and must not be used as evidence; retrieve Hauschild 2012 or the FDA medical review to establish them.]
[FLAG: extract TAFINLAR USPI §6.1 before publishing any US denominator.]
[FLAG: Yao 2015 and Yao 2017 were verified through aggregated full-text retrieval, not page-level reading — page-level re-read required before external use.]
[FLAG: Sauer 2026 full text not read; do not cite any number from it.]
[FLAG: EU Marketing Authorisation Holder for BRAFTOVI. The prior audit asserted "PIERRE FABRE MEDICAMENT (Lavaur, France) per the Braftovi EPAR Annex I." The EMA EPAR PDF would not text-extract at this data cut (returned undecoded FlateDecode streams, 890 KB binary). The document I did retrieve and read in full — the UK emc Braftovi 75 mg SmPC — names Pierre Fabre Limited. A human must open the EMA EPAR Annex I and confirm which entity is named there. Do not cite the Lavaur address until this is done.]
[FLAG: Tkacik does not test encorafenib — it names it once, in the Introduction only. It is used here solely to refute a class-exclusivity generalisation, never as evidence about encorafenib itself.]
[FLAG: genotype assignment discrepancy. The prior audit records SKMEL-30 = NRAS Q61K and SKMEL-2 = NRAS Q61R. Extraction of the Figure 7 legend returns the reverse: SKMEL-30 (BRAF-WT/NRAS Q61R) and SKMEL-2 (BRAF-WT/NRAS Q61K). Standard reference/CCLE genotypes are SK-MEL-2 = Q61R and SK-MEL-30 = Q61K, matching the prior audit and not the extracted legend. Both lines are NRAS Q61 mutants either way, so the paradox conclusion is unaffected — but a human must check the published figure before either assignment is quoted.]
[FLAG: the "and/or" construction in Biochem J 2022 means figure-level confirmation of which panels used encorafenib specifically, versus SB590885, is still required. Also: the prior audit's "who" field read "Academic cardiac-signalling group" — not a named source, which fails the audit's own rule 2. Retrieve the author list.]
[FLAG: attributed to Stuart AACR 2012 abstract 3790, full text confirmed non-retrievable — HTTP 403 — at this data cut. Not present in indexed excerpts. Unverified at primary level.]
[FLAG: the prior audit's precise hit counts could not be reproduced. It claimed "'paradox' = 0, 'dimer' = 0, 'residence time' = 0, 'dissociation' = 1 (§5.1), 'wild-type' = 2 (§4.2, §5.1)" across a "63-page EU Product Information (Annexes I–IIIB)." The EMA EPAR PDF would not text-extract at this data cut. My verification is on a different document (UK emc, Annex I equivalent) and does not cover Annexes II–IIIB or the page count. A human must re-run the term search against the EMA EPAR PDF itself before any per-term count is published.]
[FLAG: the prior audit quoted a concession, "averaged across multiple phase I-III trials conducted in heterogeneous patient populations." That phrase was not returned on extraction, and the paper has no formal limitations section. Verify verbatim or drop the quotation marks.]
[FLAG: Carr/Zager quotation not independently re-retrieved at this data cut. Verify verbatim before external use.]
[FLAG: Indini & Mandalà quotation not independently re-retrieved at this data cut.]
[FLAG: a negative result cannot be proven exhaustive. Paywalled congress archives (AACR/ASCO/ESMO, 403) and internal investor decks were not accessible. State this limitation wherever the negative is cited.]
[FLAG: this PubMed query count (encorafenib AND paradoxical, 21 records) was not independently re-executed here.]
[FLAG: denominators. FDA §5.1 states no n for any of the BRAFTOVI cutaneous-malignancy rows. The prior audit presented "n=192" for the single-agent arm as though label-tier; 192 comes from Gogas 2019 (peer), not from §5.1. A label-extraction pass returned candidate denominators (single-agent 194, BEACON 216, PHAROS 98, BREAKWATER 232 / 71) as a derived table, not verbatim §5.1 text. Every denominator in this section must be re-confirmed against USPI §6.1 / §14 before external use.]
[FLAG: Gogas full text returned HTTP 410 at this data cut; these figures are carried forward from the prior audit unverified.]
[FLAG: I do NOT adopt the ~26% COLUMBUS vemurafenib-arm number. Lancet Oncology (Dummer 2018, PMID 29573941) and the JCO COLUMBUS 5-year update (10.1200/JCO.21.02659) both returned HTTP 403/cookie walls to me. A human with journal access must retrieve the arm-by-arm cutaneous-malignancy table. Recorded as NOT RETRIEVED HERE, not as unknown — rule 4 requires an honest unknown to be a verified absence.]
[FLAG: re-establish or drop the IUPHAR ID 13808 for claturafenib.]
[FLAG: confirm claturafenib crystallography refinement software against Supplementary Methods.]
[FLAG: the analyst inference that claturafenib's orthosteric, αC-OUT, A-loop-ordered pose is the signature usually classed type I½ / αC-OUT (type IIA) is not a source claim. No source makes this assignment. Recorded as UNASSIGNED. Do not let this inference travel without this disclaimer.]
[FLAG: open claturafenib Supplementary Fig. S1A before any external use — the wild-type dose-response resides there and was not inspected, so "magnitude not reported" is softened to "not inspected".]
[FLAG: ascopubs.org returns HTTP 403 — re-confirmed 1 Sep 2026. All nine percentages from ASCO 2026 Abstract 9512 were independently reproduced from secondary indexing and are almost certainly correct, but the abstract remains UNREAD IN ORIGINAL. Do not promote to verified without obtaining the PDF.]
[FLAG: Pfizer Oncology Development molecule page HTTP 403 on direct fetch, re-confirmed. Read only via search-index snippets — INCOMPLETELY VERIFIED. Search-index snippets blend adjacent sources; per-sentence attribution between the 403 Pfizer page and the blog is unreliable.]
[FLAG: if the claturafenib "paradox breaker" class label is used downstream, it must carry "internal Fore attribution, unmirrored in the public record."]
[FLAG: the mechanistic premise that a concurrent MEK inhibitor independently suppresses paradox-driven cutaneous neoplasia is correct field knowledge but is uncited in this audit — attach a peer citation or mark it as background.]
[FLAG: the claturafenib monotherapy denominator of 18 is derived by arithmetic from the dose-level string; it sums exactly to 30, but confirm against Table 1 before external use.]
[FLAG: search the PF-07799933 / ARRY-440 patent family before the next revision — Array/Pfizer composition-of-matter and method-of-use filings routinely contain BRAF-wild-type pERK and paradox data that never reach a journal.]
[FLAG: verify claturafenib cell-line provenance line-by-line in the Supplementary Table — the panel is 19 cancer cell lines, not all patient-derived.]
[FLAG: Wichmann's "Compound Ia" and Bonfill-Teixidor's "compound 1a (C1a)" being the same entity is an inference — no located source states it. Shared authorship makes it near-certain, but a human must confirm before the melanoma-brain-metastasis dataset is attributed to mosperafenib.]
[FLAG: EMA absence for mosperafenib is not established. Record as not retrieved, not as searched and absent.]
[FLAG: the Suppl Fig S3 reading (P-ERK rebound at 24–48 h and P-c-RAF induction under mosperafenib) is my own read of band intensity. No densitometry is provided for S3, it is a single blot, and n is not stated. A human must verify against Fig 3 and the full text before this is used externally.]
[FLAG: the mosperafenib patent family member that actually claims the molecule was not identified. A patent-family search (INPADOC, on the excluded structure or on CAS 2649372-20-1) is the next step, and that document — not WO2021116050A1 — would carry any per-compound paradox data for this molecule.]
[FLAG: all Suppl Fig S2 fold-changes were read from bar heights; the numeric data are unpublished and the caption states no n, no replicate structure and no statistics. Direction and rough magnitude are unambiguous; precise values are approximate. The S3 blot read is qualitative.]
[FLAG: the JCO letter/reply exchange on mosperafenib was not read. Confirmed to have no abstract in Europe PMC; both Cloudflare-blocked. It cannot be excluded that safety or paradox is discussed. Only independent peer-reviewed challenge on record.]
[FLAG: ESMO-family abstract 126P (Ann Oncol 2026) not read — HTTP 403.]
[FLAG: confirm the mosperafenib pipeline removal against the Roche H1 2026 results presentation and the next quarterly pipeline update, and check whether any out-licensing or partnering followed.]
[FLAG: confirm the defined "Territory" of the Nuvectis/Haisco licence against the executed licence agreement or the Nuvectis 8-K exhibit before any competitive-geography conclusion. The 8-K exhibit 99.1 (accession 0001104659-26-090094) returned HTTP 403 and remains unaudited.]
[FLAG: the NXP200 PCNST ORR moves 62.5% (5/8) → 40.9% (9/22) as the evaluable denominator grows. This is an ordinary small-denominator effect, but the company-tier ">40% response rate" framing is anchored to the larger, lower number while the earlier, higher number was itself promoted at SNO. A human should decide how to characterise the trajectory.]
[FLAG: the NXP200 HGG 36.8% includes unconfirmed ("preliminary") partial responses — LB405's own wording. Only one response in the entire dataset is described as confirmed (the single CR).]
[FLAG: the NXP200 toxicity-reduction sentence ("paradox breaking properties… while reducing toxicities") could not be located in GlobeNewswire release 3338215 on retrieval. Source attribution unverified — a human must locate the original.]
[FLAG: figures, blots, assay methods and full AE tables live in the NXP200 posters/slides, not the abstracts. Every "method not named" and "no data shown" finding here is a statement about the abstract text only.]
[FLAG: confirm current Day One HQ city — the ASCO 2022 poster affiliation reads South San Francisco, CA; recent filings may read Brisbane, CA.]
[FLAG: the accessdata.fda.gov PDF path cited in the prior tovorafenib audit returns HTTP 404 — retrieve via DailyMed.]
[FLAG: tovorafenib crystal resolutions "2.75–3.5 Å" not verified at source.]
[FLAG: TAK-632 WT BRAF 8.3 nM and CRAF 1.4 nM not verified at source.]
[FLAG: verify the Tkacik cooperativity sentence verbatim — the retrieved text reads "...binding of inhibitor to the active site of one protomer increases affinity for inhibitor binding to the second protomer in the RAF dimer", which differs slightly from the prior audit's quotation.]
[FLAG: a human must reconcile which endpoint the tovorafenib NF1 GEMM 2/12 figure actually refers to — tumour volume (labels), tumour number (CHMP p.44), or nine of 48 proximal nerve volume measurements (Rastogi 2025).]
[FLAG: the Kilburn 2024 introduction sentence and its reference number could not be verified verbatim — the article is paywalled. The prior audit's attribution to "ref 25 = Sun et al. 2017" is unverified and must be checked against the published text.]
[FLAG: verify verbatim the Rastogi sentence "Tovorafenib does not result in paradoxical activation of the MAPK signaling in tumors harboring BRAF fusions" — retrieval returned it in heading-like form.]
[FLAG: the prior audit's "32 NRAS-mutant" total and its "Q2D pre-treated n=1" NRAS patient were not located in the retrieved Rasco 2023 full text.]
[FLAG: retrieve the Bessler 2024 Global NF Conference poster — it has never been retrieved at source by anyone in this audit chain.]
[FLAG: a human should use SmPC 5.2 steady-state Cmax 6.9 µg/mL and AUC 508 µg·h/mL, and SmPC 5.3's hERG IC50 of 8.9 µM described as 32-fold above the clinical plasma unbound concentration, to test whether 0.01–0.1 µM is subtherapeutic in skin — which no public document does.]
[FLAG: the FDA regulatory record for tovorafenib was never retrieved. Missing: the FDA Multi-Discipline Review for NDA 217700 (Drugs@FDA), and Singh S, Bradford D, Chatterjee S, Li X, et al. "FDA Approval Summary: Tovorafenib for Relapsed or Refractory BRAF-Altered Pediatric Low-Grade Glioma." Clin Cancer Res 2025;31(8):1383–1389, PMID 39808502 — FDA-authored and peer-reviewed. Until these are read, "the US regulator made no paradox statement" is silence, not verified absence. Only the US label has been verified.]
[FLAG: confirm the Clayton et al. eLife 95334 Version of Record publication date — one retrieval reported 13 February 2025, another 14 January 2025.]

CCorrections applied 133

What the first pass got wrong, and what the refutation pass changed.

[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] KAROULIA 2016 IS FREELY AVAILABLE AND WAS NOT A DEAD END. The audit recorded the alphaC-OUT / negative-cooperativity mechanism as review-tier only, verdict 'unknown', on the stated basis that the primary text was not read. Karoulia Z, Wu Y, Ahmed TA, et al., Cancer Cell 2016;30(3):485-498, PMID 27523909, DOI 10.1016/j.ccell.2016.06.024, is free at PMC5021590 and was retrieved and read for this review. This was a retrieval failure reported as an evidentiary absence — the exact error the audit's own rules forbid. Venue corrected from 'review' to 'peer'.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE R506 SALT-BRIDGE CLAIM IS NOT IN THE PRIMARY SOURCE AND MUST BE STRUCK. Karoulia 2016 states verbatim: 'The side-chain conformation of R506 is in a further OUT position in PB and VEM-bound structures, whereas alphaC-IN inhibitors stabilize R506 closer to IN position.' No salt bridge is mentioned anywhere in connection with R506. The formulation the audit relayed — 'displaces R506 so as to disrupt the salt bridge stabilising the dimer interface' — is an embellishment introduced by the review-tier relay (Cope N, et al., Br J Cancer 2018, PMC5765234) and attributed by the audit to Karoulia. A review was allowed to do a peer source's job. Corrected to the primary's actual wording.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] KAROULIA 2016 CARRIES AN ERRATUM THE AUDIT NEVER CHECKED: Cancer Cell 2016;30(3):501-503, DOI 10.1016/j.ccell.2016.08.008. Its content was not assessed here. [FLAG]
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE 'INDEPENDENT REPLICATION' OF HATZIVASSILIOU AND HEIDORN DID NOT HAPPEN — ONLY PROTOCOLS WERE PUBLISHED. The audit wrote that Hatzivassiliou 2010 was 'subsequently subjected to independent replication under the Reproducibility Project: Cancer Biology' and called this 'an unusually strong external check'. What exists is a Registered Report — a pre-registered study PLAN. Bhargava A, et al., eLife 2016;5:e09976, PMID 26882073, PMC4769155, states: 'This Registered Report describes the proposed replication plan... the results of the replications will be published by eLife' (future tense). Same for Heidorn: Bhargava A, et al., eLife 2016;5:e11999, PMID 26885666, PMC4769162. No completed Replication Study was located for either. The replication OUTCOME is unknown.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] DATE ERROR ON THE HATZIVASSILIOU REGISTERED REPORT: the audit gives 'eLife 2015'. It is eLife 2016 Feb 16 (PMC4769155 is correct).
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] PDB 3OG7 ASYMMETRIC UNIT CONTAINS TWO CHAINS, NOT ONE. The audit states 'asymmetric unit contains a monomer (C1 symmetry, A1 stoichiometry)'. The RCSB REST API (data.rcsb.org/rest/v1/core/entry/3OG7) returns deposited_polymer_entity_instance_count = 2 (auth chains A and B) and assembly_count = 2. The correct statement is: two protein chains in the ASU, resolving into two biological assemblies each annotated monomeric (C1 / A1). The audit conflated 'each assembly is a monomer' with 'the ASU is a monomer'.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] PDB 3OG7 IS AN ENGINEERED CONSTRUCT, NOT WILD-TYPE-BACKBONE BRAF. The RCSB polymer entity is described as 'AKAP9-BRAF fusion protein' and carries at least 15 engineered surface/solubilising mutations: I544A, I551K, Q562R, L588N, K630S, F667E, Y673S, A688R, L706S, Q709R, S713E, L716E, S720E. The audit presented 3OG7 as a plain 'BRAF V600E in complex with PLX4032' co-crystal. The residues load-bearing for the audit's mechanism narrative (L505, R506, R509) are NOT among the engineered mutations, so those contact claims survive — but the construct caveat must be stated before the structure is used to reason about native dimer behaviour.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE YAO 2015 CONCENTRATION FIGURES ARE AN ESTIMATE THE AUDIT INVENTED. The audit wrote 'monomer ~100-300 nM vs dimer >3,000 nM in the p61 splice-variant system'. Only the ratio is in the source. The verbatim sentence retrieved from PMC4894664 is: 'more than 30-fold higher concentrations of vemurafenib were required to inhibit ERK driven by p61 V600E dimers than by p61 V600E R509H monomers' (Figure 4A, cellular experiment in transfected SKBR3 cells). The absolute nanomolar values were not located and must be removed under the audit's own rule 4 (never estimate).
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE SIEVERT 2013 QUOTATION IS NOT VERBATIM AND APPEARS FABRICATED. The audit attributes to Sievert the phrase: a second-generation inhibitor that 'unlike vemurafenib, does not induce activation of wild-type BRAF'. Retrieval of PMC3625308 returned all three sentences in the paper containing 'vemurafenib'; none matches. The actual text is: 'Whereas both vemurafenib and [comparator] display nearly identical, potent inhibition of phospho-ERK (pERK) in V600EBRAF-mutant cells, the well-described paradoxical activation associated with mutant RAS expression under vemurafenib treatment is absent upon treatment with [comparator].' Quotation corrected. [FLAG: human must confirm the audit's original string does not appear in a figure legend or supplement.]
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE AACR 2012 ABSTRACT 3510 CONTENT IS RETRIEVABLE — 'NONE VERIFIED' WAS A RETRIEVAL FAILURE, NOT AN ABSENCE. The audit recorded 'The AACR full text returned HTTP 403... Only the title is established. No models, endpoints, effect sizes or author affiliations were obtained.' Substantive content is publicly indexed: co-immunoprecipitation and kinase assays implicating C-Raf in vemurafenib-induced paradoxical MAPK activation in T cells; in vivo C57BL6 mice dosed 3 weeks showing skewing of CD3+ cells toward a central-memory-like phenotype (CD44+, CD62L+, LY-6C+). Cancer Res 2012;72(8_Supplement):3510.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] A PEER-TIER PAPER MAKING THE BENEFICIAL-PARADOX CLAIM EXISTS AND THE AUDIT MISSED IT ENTIRELY. Koya RC, Mok S, Otte N, et al., 'BRAF inhibitor vemurafenib improves the antitumor activity of adoptive cell immunotherapy', Cancer Res 2012;72(16):3928-3937, PMID 22693252, PMC3422880, DOI 10.1158/0008-5472.CAN-11-2837. Abstract verbatim: 'vemurafenib paradoxically increased mitogen-activated protein kinase (MAPK) signaling, in vivo cytotoxic activity, and intratumoral cytokine secretion by adoptively transferred cells.' Vemurafenib itself, in vivo, peer venue. The audit's verdict of 'unknown' on the beneficial-consequence claim is upgraded to 'supported'.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] SU 2012 DEMONSTRATED 'ACCELERATION NOT INITIATION' EXPERIMENTALLY — THE AUDIT WRONGLY CALLED IT UNDEMONSTRATED. The audit states the acceleration-not-initiation reading 'is itself mechanistic interpretation, not a demonstrated finding'. The Su 2012 abstract states: 'In a mouse model of HRAS Q61L-mediated skin carcinogenesis, the vemurafenib analogue PLX4720 was not an initiator or a promoter of carcinogenesis but accelerated growth of the lesions harboring HRAS mutations, and this growth was blocked by concomitant treatment with a MEK inhibitor.' It is an in vivo experimental finding. Carry-over caveat: the compound was PLX4720, not vemurafenib.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE MEK-MITIGATION EVIDENCE IS STRONGER THAN THE AUDIT ALLOWED. The audit rests MEK mitigation solely on the COTELLIC label incidence figures and calls the mechanism an unsupported reader interpretation. Su 2012 additionally shows MEK-inhibitor blockade of PLX4720-accelerated HRAS-mutant lesion growth in vivo. The clinical incidence signal and an in vivo mechanistic experiment are separate strands.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] POULIKAKOS IC50 VALUES NEED THE 'PURIFIED CATALYTIC DOMAIN' QUALIFIER. Verbatim from PMC3178447: 'PLX4032 inhibited...purified catalytic domains of BRAFV600E, wild-type BRAF and CRAF (IC50s: 35, 110 and 48nM)'. The audit presented these as bare biochemical IC50s while elsewhere asserting that no purified-protein data exists. Correct framing: purified catalytic-domain (monomer) enzymology exists; purified DIMER enzymology does not.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] USPI §5.1 CONTAINS A MALIGNANCY SIGNAL THE AUDIT OMITTED: 'Cases of myeloid neoplasms amongst patients with ECD have been observed, including in patients who have received ZELBORAF. Monitoring complete blood count in ECD patients with co-existing myeloid malignancies is recommended.' (Verified verbatim, Revised 07/2026.)
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE LABEL DOES JOIN MECHANISM TO A PRESCRIBING RESTRICTION — THE AUDIT'S 'NEVER EXPLICITLY JOINS' IS TOO STRONG. USPI §1.1 Limitation of Use, verified verbatim: 'ZELBORAF is not indicated for treatment of patients with wild-type BRAF melanoma [see Warnings and Precautions (5.2)].' The paradox warning is cross-referenced from the indication itself. The audit's point survives only in its narrow form: the label does not join §5.2 to the 24% cuSCC incidence in §5.1.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE COTELLIC LABEL DISCLOSES A NON-CUTANEOUS MALIGNANCY RATE, PARTLY RESOLVING ONE OF THE AUDIT'S 'UNKNOWNS'. COTELLIC USPI §5.1, Non-Cutaneous Malignancies, verified verbatim: 'Based on its mechanism of action, vemurafenib may promote growth and development of malignancies [refer to the Full Prescribing Information for vemurafenib]. In Trial 1, 0.8% of patients in the COTELLIC with vemurafenib arm and 1.2% of patients in the vemurafenib arm developed non-cutaneous malignancies.' The audit recorded flatly that no non-cuSCC rate is disclosed at label level.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] EMA SmPC: CMML AND PANCREATIC ADENOCARCINOMA ARE TABULATED ADRs WITH AN ASSIGNED FREQUENCY, NOT MERE FOOTNOTES. Both appear in the §4.8 tabulated adverse-reaction table under frequency 'Rare' (chronic myelomonocytic leukaemia(2)(4); pancreatic adenocarcinoma(5)), with footnotes '(4) Progression of pre-existing chronic myelomonocytic leukaemia with NRAS mutation' and '(5) Progression of pre-existing pancreatic adenocarcinoma with KRAS mutation'. This is stronger label grounding than the audit claimed.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] EMA SmPC §4.8 ALSO REPORTS THAT 43% OF EXCISED LESIONS WERE BENIGN — A MATERIAL LIMIT THE AUDIT DROPPED. Verbatim: 'Most lesions classified as "other" (43%) were benign skin lesions (e.g. verruca vulgaris, actinic keratosis, benign keratosis, cyst/benign cyst).' The audit quoted the 52% keratoacanthoma figure and omitted its companion.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE USPI REVISION-DATE 'CONFLICT' IS PARTLY EXPLAINED. 'Revised: 07/2026' is confirmed in the Highlights of the gene.com PI. A second stamp, 'Revised: 11/2017', sits on the appended Medication Guide (patient labeling), which is separately dated and is not a contradiction. The DailyMed 'Revised: 5/2020' discrepancy remains unresolved. [FLAG]
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] CALLAHAN CITATION COMPLETED: Callahan MK, Rampal R, Harding JJ, et al. N Engl J Med 2012;367(24):2316-2321. PMID 23134356. DOI 10.1056/NEJMoa1208958. The audit gave no PMID, volume or pages.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] MISSING/INCOMPLETE IDENTIFIERS SUPPLIED: Hatzivassiliou 2010 DOI 10.1038/nature08833; Heidorn 2010 DOI 10.1016/j.cell.2009.12.040; Halaban 2010 Pigment Cell Melanoma Res 2010;23(2):190-200, DOI 10.1111/j.1755-148X.2010.00685.x; Yao 2017 PMID 28783719, PMC5648058; Bollag 2010 DOI 10.1038/nature09454; Karoulia 2016 PMID 27523909, PMC5021590.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] THE USPI NEVER SAYS 'BRIM-3'. It says only 'Trial 1'. The audit's 'Trial 1 (BRIM-3)' is an external mapping (EMA calls it NO25026), not a label statement. Minor, but it is an unsourced identity assertion presented inside a label-tier citation.
[Vemurafenib (ZELBORAF). Data cut: 1 September 2026.] SPONSOR AUTHORSHIP AND FUNDING WERE NOT DISCLOSED FOR THE AUDIT'S STRONGEST PEER CITATIONS. Su 2012: first author affiliation Hoffmann-La Roche, Nutley NJ; Plexxikon co-authors (Bollag, Zhang, Habets, Burton, Wong, Nguyen); 'Funded by Hoffmann-La Roche and others'. Poulikakos 2010: Bollag (Plexxikon) co-author. Karoulia 2016: C. Zhang and G. Bollag are Plexxikon employees (declared). These remain peer tier, but the audit flagged Hatzivassiliou's Genentech provenance while staying silent on three others.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] MISQUOTATION, Karoulia 2016 (highest-stakes quote in the file). Audit renders: 'DAB and VEM showed potent inhibition of monomeric BRAF-V600E but 30 to 100-fold less potency for inhibition of dimeric BRAF-V600E.' Source (Cancer Cell 2016;30:485-498, verified at PMC5021590) reads: 'PB, DAB and VEM showed potent inhibition of monomeric BRAFV600E but 30 to 100-fold less potency for inhibition of dimeric BRAFV600E.' The audit deleted the leading compound (PB = PLX7904). Material consequence: the 30-100x dimer potency loss is a shared property of the compared alphaC-OUT compounds in that experiment, not a dabrafenib-distinguishing finding, and the audit's framing implies otherwise.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] MISIDENTIFIED COMPOUND. Audit writes 'AZ-vemurafenib'. No such agent exists. The paper's compound is 'AZ-VEM', a chimeric tool compound described as 'an AZ-like core structure attached to the same substituted sulphonamide group from VEM'. It is not vemurafenib, not an AstraZeneca clinical asset, and not a marketed drug. Corrected throughout.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] TRUNCATED QUOTE, Karoulia 2016. Audit: dabrafenib and [AZ-VEM] 'promote RAF priming and dimerization more potently than VEM ... and at levels closer to alphaC-IN inhibitors'. Source: 'DAB and AZ-VEM, although alphaC-OUT inhibitors, promote RAF priming and dimerization more potently than VEM and PB and at levels closer to alphaC-IN inhibitors.' The audit dropped 'and PB'.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] MISQUOTATION THAT INVERTS THE SOURCE, Holderfield 2014 (Br J Cancer). Audit quotes as a flat assertion: 'The lower incidence of cutaneous lesions seen with dabrafenib is owing to higher potency against BRAF-V600E...'. Source (verified at PMC4134487) reads: 'It has been speculated that the lower incidence of cutaneous lesions seen with dabrafenib is owing to higher potency against BRAFV600E compared with WT BRAF and CRAF, whereas vemurafenib may be relatively equipotent.' The audit deleted 'It has been speculated that', then criticised the review for asserting what it had in fact explicitly flagged as speculation.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] AFFILIATION ERROR. Audit: 'Holderfield, Nagel, Stuart (review authors; Novartis-affiliated at time of writing)'. As printed: M Holderfield, UCSF Helen Diller Family Comprehensive Cancer Center, San Francisco; T E Nagel and D D Stuart, Novartis Institutes for Biomedical Research, Emeryville. The first author was NOT Novartis-affiliated.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] FALSE RETRIEVAL BARRIER. Audit records the Br J Cancer 2014 review as 'full text not opened (publisher auth wall); sentence verified only via indexed search snippet'. Free full text is open at PMC4134487. The 'unknown' was not a gap in the record, it was an unattempted retrieval — and retrieving it changed the finding (see hedge correction above).
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] MEASUREMENT MIS-FRAMED, Yao 2015. Audit states 'the relative affinity for the first protomer site is ~30-fold higher than for the second once the first is drug-occupied'. What was measured: 'More than 30-fold higher concentrations of vemurafenib were required to inhibit ERK driven by p61 V600E dimers than by p61 V600E R509H monomers' — a cellular ERK-inhibition potency shift, not a binding-affinity ratio between protomer sites. The audit converted a functional potency ratio into a biophysical affinity statement.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] UNVERIFIED AND MECHANISTICALLY INVERTED QUOTE, Yao 2017. Audit quotes class 3 mutants as able to 'trigger paradoxical ERK pathway activation by binding to a drug-free protomer'. This could not be confirmed as verbatim Nature 2017 text, and as written it is backwards: paradoxical activation arises when drug binds ONE protomer and transactivates the DRUG-FREE one. Quote removed pending page-level verification.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] VENUE MIS-TAG. COMBI-v claim is tagged venue 'conference'. The 1% vs 18% cuSCC/keratoacanthoma figure quoted comes from Robert C et al. NEJM 2015;372:30-39 (peer, verified: 'Cutaneous squamous-cell carcinoma and keratoacanthoma occurred in 1% of patients in the combination-therapy group and 18% of those in the vemurafenib group') and is corroborated at label tier by the EMA SmPC. Retagged 'peer'.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] TRIAL POPULATION MIS-STATED. Audit: 'n=247 randomised, 162 to combination arms.' Verified abstract: 247 patients enrolled overall; 85 in the phase 1 pharmacokinetic/safety portion; 162 then randomised across dabrafenib 150 + trametinib 1, dabrafenib 150 + trametinib 2, AND dabrafenib monotherapy. 162 were not all assigned to combination.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] PDB 4XV2 WRONGLY LEFT UNVERIFIED. RCSB entry 4XV2 = 'B-Raf Kinase V600E oncogenic mutant in complex with Dabrafenib', ligand P06 (dabrafenib), biological assembly a homodimer (C2, A2) with dabrafenib bound in both chains. This is the V600E dabrafenib dimer structure and is the most directly relevant of the four; the audit relegated it to 'cited in secondary sources — unverified'.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] MISSED LABEL-TIER PARADOX CLAIM. TAFINLAR USPI 3/2026 Section 1.8 Limitations of Use: 'TAFINLAR is not indicated for treatment of patients with wild-type BRAF solid tumors [see Warnings and Precautions (5.2)].' This is a paradox-driven negative indication written into the label and cross-referenced to the tumor-promotion warning. The audit's paradox_claims list omits it entirely.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] 'COMPANY TIER IS EMPTY' IS WRONG AS STATED. A company-tier document exists and was located: GSK press release, 29 May 2013 (FDA approval of Tafinlar and Mekinist), which restates the Section 5.2 paradox sentence verbatim and additionally reports BREAK-3 comparator figures (cuSCC in 7% of dabrafenib-treated patients vs none on chemotherapy; new primary malignant melanoma 2% (3/187) vs none on chemotherapy). Awareness tier only — it establishes nothing — but the tier is populated, and it happens to contain the very BREAK-3 numbers the audit listed as unverifiable.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] NCI-MATCH GAP MIS-DESCRIBED. Audit says sub-protocol H is 'dabrafenib+trametinib in BRAF alterations' and 'may contain class-resolved results'. EAY131-H enrolled BRAF V600(E) mutation-positive tumours only (melanoma, thyroid and colorectal excluded; NSCLC excluded after approval). It cannot contain class II or class III data. Sub-protocol R is trametinib monotherapy, so it contains no dabrafenib data at all. Neither can close the gap.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] IDENTITY INCOMPLETE. TAFINLAR capsules are NDA 202806; TAFINLAR tablets for oral suspension are a separate application, NDA 217514 (the two are approved together in current supplement letters, e.g. 202806/S-036 and 217514/S-010). EU: MAH Novartis Europharm Limited, Dublin; EU/1/13/865; date of first authorisation 26 August 2013. None of NDA 217514, the EU MA number or the EU authorisation date appears in the audit.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] CITATION INCOMPLETE. Sauer et al. 2026 is: Sauer N, Giedziun P, Calik J, Wiela-Hojenska A. Cutaneous Adverse Drug Reactions Associated With BRAF and MEK Inhibitors: A Real-World Analysis of WHO Pharmacovigilance Data. Clin Pharmacol Ther. 2026;120:274-285. doi:10.1002/cpt.70326.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] UNFLAGGED SOURCE DISCREPANCY. Long 2014 NEJM reports cuSCC in 9% of the dabrafenib-monotherapy arm of COMBI-d; the EMA SmPC reports 10% (22/211) for dabrafenib monotherapy in the same study (MEK115306). The audit presents the SmPC figure as 'label-corroborated' without noting the two sources do not give the same number.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] PRECISION, King 2013. Audit gives 'Cellular pERK/pMEK inhibition 3-6 nM'. Source: 'In cell lines encoding BRAFV600E, dabrafenib inhibited pERK and pMEK in a concentration-dependent manner with IC50 values of 3 nM and 6 nM, respectively' — pERK 3 nM, pMEK 6 nM, not a 3-6 nM range.
[Dabrafenib (TAFINLAR). Data cut: 1 September 2026.] KAROULIA PANEL INCOMPLETE. The audit lists the eight profiled inhibitors correctly (AZ-628, TAK-632, LY3009120, GDC-0879, SB-590885, PLX7904, vemurafenib, dabrafenib) but omits the chimeric AZ-VEM used in the structural/priming comparison from which the dabrafenib R506 conclusion is drawn.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] IDENTITY — FABRICATED CODE NUMBERS. The audit lists 'ARRY-085 / ARRY-424704 (Array codes appearing in some filings)' as encorafenib aliases. Neither is verifiable. ChEMBL CHEMBL3301612 lists exactly five research codes: LGX-818, LGX818, NVP-LGX818, NVP-LGX-818-NXA, NVP-LGX818-NXA. Array BioPharma's own acquisition press release is titled 'Array Announces Agreement To Acquire ENCORAFENIB (LGX818)' and carries no ARRY code for the molecule. STRIKE both. Correct additions: CAS 1269440-17-6; first approval 2018. Source: ChEMBL API record CHEMBL3301612; investor.arraybiopharma.com.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] IDENTITY — MARKETING AUTHORISATION HOLDER NOT VERIFIED AS STATED. The audit asserts 'PIERRE FABRE MEDICAMENT (Lavaur, France) per the Braftovi EPAR Annex I'. The EMA EPAR PDF would not text-extract at this data cut (returned undecoded FlateDecode streams). The document I did retrieve and read in full — the UK emc Braftovi 75 mg SmPC — names the MAH as 'Pierre Fabre Limited'. The EPAR Annex I attribution is unverified.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] VENUE MIS-ATTRIBUTION INSIDE THE LABEL TIER. The audit's behaviour_by_alteration states 'Label states inhibition of V600E, V600D and V600K.' That sentence is the EU SmPC §5.1 ('Encorafenib suppresses the RAF/MEK/ERK pathway in tumour cells expressing several mutated forms of BRAF kinase (V600E, D and K)'), NOT the FDA label. FDA USPI §12.1, retrieved verbatim, names only BRAF V600E, wild-type BRAF and CRAF. Correct the jurisdiction.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] VENUE INTEGRITY — A SEARCH RECORD TAGGED AS PEER. Claim 18 (the systematic negative: no source calls encorafenib a paradox breaker) is tagged venue:'peer'. It is an audit search record. Peer papers are cited inside it, but they support the positioning sub-claim, not the exhaustive negative. Retag venue:'audit/search record'; the negative cannot inherit peer tier from its constituents.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] ABSENCE CLAIM PARTLY UNVERIFIED AT SOURCE. Claim 4's precise hit counts ('paradox'=0, 'dimer'=0, 'residence time'=0, 'dissociation'=1, 'wild-type'=2, across a '63-page EU Product Information, Annexes I–IIIB') could not be reproduced: the EMA EPAR PDF was not text-extractable at this data cut. I independently verified 'paradox' / 'paradoxical' / 'dimer' = 0 in a DIFFERENT document — the UK emc Braftovi 75 mg SmPC (Annex I equivalent). The core absence survives; the per-term counts and the annex coverage do not.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] VERDICT-FIELD CATEGORY ERROR. Claims 4 and 18 carry verdict 'contradicted' while every other entry uses the verdict to mean 'is the claim supported by its evidence'. Both claims are, on the evidence, SUPPORTED (claim 4 as an absence I partially re-verified; claim 18 as a negative search result). The audit is using 'contradicted' to mean 'contradicts a paradox-sparing narrative'. Two different axes are collapsed into one field. Restated below as verdict + separate 'bearing on paradox-sparing narrative'.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] COTTO-ROIS CELL-LINE GENOTYPES SWAPPED. The audit records SKMEL-30 (NRAS Q61K) and SKMEL-2 (NRAS Q61R). Extraction from the Nat Commun Figure 7 legend returns SKMEL-30 (BRAF-WT/NRAS Q61R) and SKMEL-2 (BRAF-WT/NRAS Q61K) — the reverse. Standard CCLE/reference genotypes are SK-MEL-2 = NRAS Q61R and SK-MEL-30 = NRAS Q61K, which matches the audit and not the extracted legend. One of the two is wrong; both lines are NRAS Q61 mutants either way, so the paradox conclusion is unaffected. Flagged for human verification against the published figure.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] CLAYTON ET AL. CITED AS AN UNPUBLISHED PREPRINT. The audit cites 'bioRxiv 2023.12.12.571293 / eLife 95334'. It is published: Clayton J, Romany A, Matenoglou E, Gavathiotis E, Poulikakos PI, Shen J. Mechanism of dimer selectivity and binding cooperativity of BRAF inhibitors. eLife 2025;13:RP95334, PMID 39945510. The audit's substantive point is CONFIRMED — 'encorafenib' appears exactly once, in the Introduction, with no IC50, cooperativity or dimer-selectivity value.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] MARINI POSE MISDESCRIBED AS HOMOLOGY MODELLING. The audit says 'the Marini homology-model work'. Marini et al. performed DOCKING onto an experimental BRAF–dabrafenib co-crystal (PDB 5CSW, 2.66 Å) after self-docking validation of dabrafenib. It is a docked pose on a real template, not a homology model. Also, the two surrogate structures were rejected for DIFFERENT reasons the audit merged: 5CSX rejected as a DFG-out crystallisation artefact; 6UUO (3.29 Å) rejected for a disordered P-loop and activation loop.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] MARINI CONTACT RESIDUES PARAPHRASED BEYOND SOURCE. The audit states 'hydrogen bond from the sulfonamide to Lys483; pyrimidinyl interaction with the Cys532 backbone nitrogen.' Verbatim source: 'hydrogen bonds between residues Asp594 and Phe595 (DFG segment), Lys483 and the sulfonylamide moiety' and 'a typical 1–3 hydrogen bond with the backbone of Cys532 in the hinge region'. The 'pyrimidinyl' attribution is not in the retrieved text.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] RESIDENCE-TIME QUALIFIER DROPPED — MATERIAL TO PARADOX. Koelblinger 2018 verbatim: encorafenib has 'a substantially increased dissociation half-life (T1/2-diss) FROM V600E-MUTANT BRAF of more than 30 h, as compared with 2 and 0.5 h reported for dabrafenib and vemurafenib.' The >30 h figure is measured on the MUTANT target. It carries no information about dwell time on wild-type BRAF or CRAF — which is where paradox occurs. The audit reproduced the number without the qualifier that voids most residence-time-to-paradox reasoning in either direction.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] KOELBLINGER cuSCC FIGURE AND ITS PROVENANCE MISSTATED. The audit writes 'the figure most often quoted... is 3.7% (Adelmann 2016) or 4% (Koelblinger 2018)' as if interchangeable. Verified: Adelmann's 3.7% sits in a sentence citing a multi-reference block [3],[7]–[14]; Koelblinger's 4% is cited to reference [7] = Delord 2017, the phase I dose-escalation paper. Different provenance, different denominators. Both remain below the label's 8%.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] STUART AACR 2012 — THREE OF FOUR ATTRIBUTED FACTS ARE NOT IN THE RETRIEVABLE ABSTRACT. Abstract number 3790 is CONFIRMED (Cancer Res 2012;72(8_Suppl):3790) and AACR full text is CONFIRMED 403 at this data cut. Indexed abstract text confirms only: tumour regression in BRAF-mutant xenografts at doses as low as 1 mg/kg, and 'LGX818 was inactive against BRAF wild-type tumors at doses up to 300 mg/kg bid'. It does NOT contain '>400 cell lines expressing wild-type BRAF', the >30 h dissociation half-life, or gastric hyperplasia. The audit sourced all four to this abstract; three are secondary attributions via review paraphrase.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] COLUMBUS SINGLE-AGENT DENOMINATOR IS IMPORTED, NOT LABEL-STATED. FDA §5.1 gives the single-agent figures (cuSCC/KA 8%, BCC 1%, new primary melanoma 5%) with NO n in that section. The audit presents 'n=192' as if label-tier. The 192 comes from Gogas 2019 (peer). Same issue for the BEACON, PHAROS and BREAKWATER denominators the audit implies.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] EU DERMATOLOGIC SURVEILLANCE SCHEDULE NOT FOUND WHERE CITED. The audit attributes 'dermatologic evaluation before treatment, every 2 months on treatment, and for up to 6 months after discontinuation' to both FDA §5.1 and EU §4.4. Verified verbatim in FDA §5.1. NOT present in the retrieved EU §4.4 cutaneous/non-cutaneous malignancy paragraphs, which instead specify head/neck exam, chest/abdomen CT, anal and pelvic examination and CBC. Do not cite EU §4.4 for the dermatologic schedule without re-checking.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] BIOCHEM J CITATION INCOMPLETE AND AUTHORS UNNAMED. Correct: Biochem J. 2022;479(3):401–424, PMID 35147166. The audit's 'who' field reads 'Academic cardiac-signalling group', which is not a named source and fails the audit's own rule 2. Substance CONFIRMED: 'The experimental Type 1 RAF inhibitor, SB590885, and/or encorafenib... increased ERK1/2 phosphorylation in cardiomyocytes, and promoted hypertrophy, consistent with a RAF paradox effect, with both promoting cardiac hypertrophy in mouse hearts in vivo.' The 'and/or' construction means figure-level confirmation of which panels used encorafenib is still required.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] ADELMANN ASSAY TIMEPOINT INCOMPLETE AND A QUOTED LIMITATION UNVERIFIED. Verified: pERK induction primary readout at 15 min, plus a 72-h time course peaking at 6 h; antiproliferative arm in A375 at 72 h; index = pERK induction EC80 / A375 IC80. The audit's quoted concession 'averaged across multiple phase I-III trials conducted in heterogeneous patient populations' was NOT returned on extraction and the paper has no formal limitations section. The audit's substantive point (multi-reference block, no single denominator) stands on the verified citation format.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] MISSED LABEL-TIER FACT THAT STRENGTHENS THE CASE. Both labels state encorafenib inhibits BRAF V600E at 0.35 nM, WILD-TYPE BRAF at 0.47 nM and CRAF at 0.30 nM. CRAF is inhibited MORE potently than the mutant target and wild-type BRAF is near-equipotent — there is NO enzymatic selectivity for V600E over wild-type RAF at label tier. This is the direct biochemical substrate for paradox, it required no inference, and the audit reported the three numbers without drawing it. It also sits in the same EU SmPC paragraph that calls encorafenib 'potent and highly selective'.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] MISSED LABEL-TIER OFF-TARGET DATA. FDA §12.1 verbatim: 'Encorafenib was also able to bind to other kinases in vitro including JNK1, JNK2, JNK3, LIMK1, LIMK2, MEK4, and STK36 and reduce ligand binding to these kinases at clinically achievable concentrations (≤0.9 µM).' Absent from the audit; relevant to any selectivity framing.
[Encorafenib (BRAFTOVI) — adversarial re-verification of the paradox-claim audit. Data cut 1 September 2026.] RCSB SEARCH WAS ASSERTED, NOT SHOWN — NOW ACTUALLY RUN. The audit states 'Independent RCSB searching at this data cut returned no encorafenib-liganded BRAF entry.' Executed against the RCSB search API: full-text 'encorafenib' returns total_count = 1, entry 9BP8 = 'Crystal structure of BRAF kinase domain with PF-07799933', X-ray 1.73 Å, UniProt P15056, ligand A1AQY, deposited 2024-05-07, released 2025-05-14. Encorafenib is NOT the bound ligand. The conclusion SURVIVES and is now evidenced rather than asserted — but the audit's phrasing implied a null search, and the search is not null.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] MONOTHERAPY SAFETY DENOMINATOR IS 18, NOT 30. The audit's headline conclusion ("the total public monotherapy safety denominator is 30 patients with a silent skin section") is wrong. Yaeger 2024 dose-level enrolment reads: 50 mg QD (n=4 mono), 150 mg QD (n=4 mono, n=4 +binimetinib, n=4 +cetuximab), 225 mg BID (n=5 mono, n=4 +binimetinib), 450 mg BID (n=5 mono) = 18 monotherapy + 12 combination = 30 treated. The monotherapy paradox-readout denominator is therefore 18. This strengthens the audit's own conclusion but the stated number was incorrect. Source: Yaeger et al., Cancer Discov 2024;14(9):1599-1611, PMC11372368, Results.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] TRUNCATED VERBATIM QUOTE ON THE COCRYSTAL. The paper reads "...both consistent with an inactive kinase conformation that may destabilize RAF dimers OR PREVENT DIMERIZATION." The audit dropped "or prevent dimerization" from a quotation presented as verbatim. Source: PMC11372368.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] "THE TRIAL EXCLUDED RAS-MUTANT PATIENTS" IS NOT IN THE RECORD. ClinicalTrials.gov NCT05355701 (API v2, retrieved 1 Sep 2026) contains no RAS/KRAS/NRAS exclusion criterion. RAS-mutant BRAF-wild-type patients are absent because enrolment REQUIRES a qualifying BRAF alteration (V600 or non-V600 Class II/III), not because RAS mutation was an exclusion. The audit converted an inclusion-criterion consequence into an asserted exclusion criterion.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] MONOTHERAPY vs COMBINATION AE ARMS CONFLATED. "Blurred vision 22%" is a MONOTHERAPY TEAE, not part of the combination list. Correct arms: monotherapy (n=18) any-grade 94%, grade >=3 28% — fatigue 44%, headache 28%, blurred vision 22%, lipase increased 16%. Combination (n=12) any-grade 100%, grade >=3 44% — peripheral oedema 33%, acneiform rash 28%, diarrhoea 28%, fatigue 28%. Source: PMC11372368.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] PDB 9BP8 LINKAGE WAS UNDER-CLAIMED, AND THE STRONGEST CORROBORATION WAS MISSED. Depositor "T.-C. Mou" is co-author "Mou TC" on Yaeger 2024 (affiliation Pfizer Boulder Research and Development), and the 9BP8 entity is a single ~294-residue BRAF kinase-domain chain, matching the paper's BRAF-KDL construct (residues 432-726 = 295 residues). The linkage is strongly corroborated by shared authorship and construct length, though the deposit's primary citation is still "To be published" and the paper's Data Availability statement carries no accession. Sources: RCSB 9BP8; PubMed 38691346 author list.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] CRYSTALLOGRAPHY METHODS INCOMPLETE/INACCURATE. The paper describes TWO constructs — BRAF-KDL (residues 432-726) and BRAF-KDS (residues 445-723, three solubilising point mutations) — and molecular replacement used PDB 4MNE as search model via PHASER. The audit named only KDL and attributed the work to "PHENIX/COOT", which is not what the retrieved methods text states. [Verify against Supplementary Methods.]
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] QUOTE MISATTRIBUTED BETWEEN TWO CLAIMS. "Significantly sparing normal (wild-type) cells to reduce off-target side effects" is the New Drug Approvals blog's wording (claim 5), NOT Pfizer's. Pfizer's own page wording (via search index; the page returns HTTP 403) is "selective for inhibition of mutated BRAF, which may contribute to a favorable therapeutic safety profile." The audit duplicated the blog's sentence into the company-tier claim, which inflates the apparent corporate assertion.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] IUPHAR/BPS LIGAND ID 13808 IS UNVERIFIED. guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=13808 returns a login screen; the identifier could not be confirmed. The audit listed it in resolved_identity as established fact while its own citation block admitted the page was never read. Removed from confirmed identifiers.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] IN VIVO BRAF-INDEL ACTIVITY OMITTED. Yaeger 2024 reports regressions in a BRAF-indel pancreatic cancer xenograft. The audit characterised indels only by the weaker cell pERK IC50 (113-179 nmol/L) and recorded no in vivo indel data.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] PANEL DESCRIPTOR OVERSTATED. The pERK panel is 19 cancer cell lines, not all patient-derived — MEL21514 is patient-biopsy-derived, others are established repository lines (e.g. A375, HT29). "19 patient-derived cell lines" is not accurate. [Verify line-by-line provenance in Supplementary Table.]
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] WILD-TYPE SPARING RATIO IS A LOWER BOUND, NOT A RANGE. Wild-type pERK IC50 is reported as ">=9,800 nmol/L", so "~1,400-14,000x sparing" is a floor, not an estimate. Restate as ">=1,400x relative to the weakest Class I line and >=14,000x relative to the most potent".
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] AUTHORSHIP CHARACTERISATION IMPRECISE. The audit calls it "a single Pfizer-authored paper". It is sponsor-led with academic clinical co-authors (Yaeger, Rosen, Diamond and others at MSK; multi-site investigators). All preclinical, structural, target-engagement and co-IP work is by Pfizer Boulder R&D authors. The correct statement is "all mechanistic paradox-relevant data are sponsor-generated", which is the point that actually matters and survives the correction.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] IDENTIFIERS RE-VERIFIED AND CONFIRMED: claturafenib = PF-07799933 = ARRY-440; CAS 2754408-94-9; PubChem CID 165150001; C18H15Cl2F2N5O3S, MW 490.3; IUPAC N-[2-chloro-3-[(5-chloro-3-methyl-4-oxoquinazolin-6-yl)amino]-4-fluorophenyl]-3-fluoroazetidine-1-sulfonamide; PDB ligand A1AQY. Paper verbatim: "PF-07799933 (ARRY-440), an orthosteric, pan-mutant BRAF inhibitor, was discovered by Array BioPharma." PF-07799544 = polfurmetinib confirmed. Identity section is sound.
[claturafenib (PF-07799933 / ARRY-440) — adversarial re-verification, data cut 1 September 2026] LABEL-TIER ABSENCE RE-VERIFIED BY RETRIEVAL, NOT BY SILENCE: PubMed returns exactly 2 records for PF-07799933/ARRY-440/claturafenib (Yaeger 2024; Scotece & Drosten 2024); no Drugs@FDA or EMA entry exists; every source including Pfizer's own and the aggregator blog states the molecule is investigational, Phase 1. The empty label tier is confirmed, not assumed.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] PATENT MIS-SCOPED (most damaging error). The audit treats WO2021116050A1 as the sponsor's own paradox claim about mosperafenib and calls it 'the honest version of the claim,' flagging only that Example 3 is the 4-chloro analogue. Verified at source: claim 1 of WO2021116050A1 carries a proviso expressly EXCLUDING mosperafenib — verbatim, 'with the proviso that (3R)-N-[2-cyano-4-fluoro-3-(3-methyl-4-oxo-quinazolin-6-yl)oxy-phenyl]-3-fluoro-pyrrolidine-1-sulfonamide is excluded or a pharmaceutically acceptable salt thereof' (proviso repeats in claims 1, 2, 3 and throughout the embodiments). Mosperafenib is not merely unexemplified in this patent; it is affirmatively disclaimed. The patent's 'considerably less paradoxial activation' language and its A375/HCT116 HTRF data therefore describe a follow-on series from which this molecule is carved out. Source: https://patents.google.com/patent/WO2021116050A1/en, full text retrieved and grepped 1 Sep 2026.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] PATENT ASSAY DETAIL WRONG/OMITTED. The comparator in the patent's paradox assay is DABRAFENIB (plus reference compound AR-25 from WO2012/118492), not encorafenib, and the quantified claim is that 'the maximum paradox inducing effect is substantially reduced by more than 25% for all examples' — a >=25% reduction relative to dabrafenib, i.e. every exemplified compound STILL induces a measurable paradox signal. The readout is a bell-shaped P-ERK activation curve whose EC50 is declared 'not applicable' when activation fails to reach 50% of the dabrafenib maximum.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] MISSED SPONSOR DISCLOSURE THAT REFUTES A RECORDED UNKNOWN. The audit records 'NO PHASE II/III PROGRAMME CONFIRMED... whether the asset is advancing, partnered, or shelved could not be established from public sources as of 1 September 2026.' Refuted. Roche's own Q2 2026 development-pipeline update (Roche Group development pipeline, pharma half-year 2026, status as of 23 July 2026, https://assets.roche.com/f/176343/x/cb875526bd/pharmahy26.pdf, page 2) lists 'RG6344 mosperafenib (BRAFi) - solid tumors' under the column heading 'Removed from phase I', as one of 4 NMEs (with RG6382, RG6496, RG6468). Verified by coordinate-level text extraction: the entry sits at x=49.3, y=164.2, directly beneath the 'Removed from phase I' header at x=89.4, y=209.1. RG6344 appears nowhere else in the 68-page document and is in no current phase table.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] MISSED PRIMARY EVIDENCE IN THE SAME SUPPLEMENT THE AUDIT DOWNLOADED. The audit states 'no other figure addresses paradoxical activation.' False. Suppl Fig S3 (MCT 2026; file mct-25-0562_suppl_fig_s3_suppsf3.png, already present in the audit's own scratchpad) shows HT29 (BRAF V600E) treated with mosperafenib or encorafenib at 300 nM, +/- cetuximab 5 ug/mL, +/- EGF 10 ng/mL, over 0/3/24/48 h, blotted for P-ERK, ERK, P-c-RAF, c-RAF, P-AKT. Under mosperafenib monotherapy P-ERK rebounds markedly at 24 h and 48 h, and P-c-RAF is induced at 24-48 h — an on-pathway RAF-activation readout for mosperafenib itself.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] SUPPL FIG S2 READ IS HALF THE FIGURE. Every panel of S2 contains a monotherapy block AND a +cetuximab 5 ug/mL block; the audit read only the monotherapy block. Re-read at magnification: the maximum mosperafenib pERK induction in the figure is ~3.65x DMSO in HCT116 at 100 nM WITH cetuximab (not ~2.85x as recorded). +cetuximab values (mosperafenib / encorafenib, vs the overall DMSO control): HCT116 DMSO ~2.2/~2.15, 10 nM ~3.0/~3.2, 100 nM ~3.65/~4.65, 1000 nM ~1.85/~2.6. SW480 DMSO ~1.75/~2.2, 10 nM ~1.85/~3.0, 100 nM ~1.95/~3.65, 1000 nM ~1.95/~1.85. SW48 DMSO ~0.45/~0.9, 10 nM ~0.6/~0.8, 100 nM ~0.85/~0.95, 1000 nM ~0.8/~1.0. The cetuximab combination is the clinically relevant CRC regimen and was omitted entirely.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] COMPARATIVE CLAIM OVERSTATED. The audit's 'supported' verdict rests on 'across all three BRAF wild-type lines and at essentially every concentration, the mosperafenib bar is lower than the encorafenib bar.' Verified false in SW480 at 1000 nM, in BOTH blocks: monotherapy mosperafenib ~1.15 vs encorafenib ~0.95; +cetuximab mosperafenib ~1.95 vs encorafenib ~1.85. Mosperafenib is the higher bar in 2 of 8 SW480 comparisons.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] 'VENDOR FIGURES CARRY NO PRIMARY CITATION' IS FALSE. The ProbeChem datasheet for RG6344 explicitly cites 'Wichmann J, et al. Clin Cancer Res. 2022 Feb 15;28(4):770-780' for its values. It also carries two figures the audit omitted: kinase IC50 <1.77 nM across BRAF mutants (V600E/K/A/D) and cytotoxic activity 5.2-30.2 nM across 94 cell lines. Source: https://www.probechem.com/products_RG6344.html (retrieved 1 Sep 2026).
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] ACTA PHARM SIN B TABLE NOT DECODED — AND IT CORROBORATES THE PAN-RAF PATTERN. The audit quotes '<5E-10 M and 1.77E-09 M' without identifying the targets. Column headers retrieved from the local copy of PMC11120325: 'Conformation binding to RAF | Cell-free assay (IC50: nmol/L): ARAF, BRAF, CRAF, BRAF V600E | Effects on RAF dimers (confirmed experimentally) | Preclinical effects | Phase | Ref.' The Compound Ia row therefore reads ARAF = NA, BRAF (wild-type) = <5E-10, CRAF = NA, BRAF V600E = 1.77E-09. An independent review thus tabulates the compound as MORE potent against wild-type BRAF than against BRAF V600E — concordant with the vendor Kd pattern (0.6 nM WT vs 1.2 nM V600E), and citing refs 146/147 = Bonfill-Teixidor 2022 and Wichmann 2022. The figures are traceable, not untraceable.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] 'NO RAF DIMERISATION ASSAY HAS BEEN REPORTED' IS TOO FLAT. The Acta Pharm Sin B dimer column is explicitly headed 'Effects on RAF dimers (confirmed experimentally)' and enters 'BRAF/BRAF' for Compound Ia, citing the two 2022 Roche papers. Correct statement: no dimer assay is publicly READABLE; a specialist review asserts an experimentally-confirmed BRAF/BRAF dimer effect and points to the two paywalled 2022 papers as its source.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] MISSED SOURCE — a sixth Europe PMC record. Biomolecules 2026;16:543 (PMC13114179), an independent review, states: 'Conventional BRAF inhibitors may paradoxically activate MAPK signaling in RAS-mutant or WT cells via RAF dimerization... In contrast, "paradox-breaker" BRAF inhibitors (e.g., PLX8394, PLX7904, mosperafenib) inhibit RAF dimerization and maintain MAPK suppression.' This is the ONLY source located that makes an explicit RAF-DIMERISATION claim about mosperafenib by name, and it is review tier with no supporting experiment. It must be recorded as a propagating assertion.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] EUROPE PMC CORPUS MISCOUNTED. The audit says the 6 records are '4 primary papers plus a letter/reply pair.' Correct composition: 3 named-molecule primary papers (MCT 2026, JCO 2026, Cancer Res Commun 2026) + the letter/reply pair (PMIDs 42659591 / 42659589) + 1 review (Biomolecules 2026, PMC13114179). Wichmann 2022 and Bonfill-Teixidor 2022 do not contain the string 'mosperafenib' and are not in this result set.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] ASCO 3542 AUTHORSHIP FABRICATED. The audit's 'who' field reads 'Vieito/Kotani-presented programme.' No author named Kotani exists on this abstract. Crossref author list: Fontana E (first author), Vieito Villar M, Castanon Alvarez E, Matos I, Bechter OE, Moreno I, Han CH, Eefsen RL, Pinato DJ, Plummer R, Prenen H, Schnetzler G, Pettazzoni P, Dejardin D, Abiraj K, Cinato E, Flinn N, Kratochwil NA, Roller A, Keshelava N.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] ASCO PHARMACODYNAMIC QUOTE MISATTRIBUTED. The audit states 'Every venue reports "PK-derived pERK inhibition" (>80% at the ASCO cut).' ASCO 3542 says only 'reaching Ctrough levels exceeding pERK inhibition > 80%' — the qualifier 'PK-derived' appears in AACR CT017 ('PK-derived pERK inhibition >90%') and in JCO 2026 ('sustained PK-derived pERK inhibition >=90%'), not in ASCO. The conclusion (no measured tissue pERK anywhere) stands; the quotation did not.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] IDENTITY FRAGILITY OVERSTATED. 'mosperafenib = RG6344 = RO7276389' IS peer-reviewed: MCT 2026 abstract, verbatim, 'the next-generation BRAFi mosperafenib (RG6344/RO7276389)' (also at conference tier in ASCO 3542 and AACR CT017). Only the link to 'Compound Ia' is conference-only. Further, Cancer Res Commun 2026 (open access, full text read) states 'Mosperafenib is a novel, brain-penetrant BRAFi designed to overcome this paradox, showing promising activity in preclinical models (6)' — where its reference 6 is Wichmann, Clin Cancer Res 2022. A peer-reviewed paper therefore attributes mosperafenib's preclinical characterization to the Wichmann paper.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] SUPPL FIG S1 INCOMPLETE AND ONE COMPARISON WRONG. Full table: LS411N mosperafenib IC50 10.8 / IC80 31.87 / IC95 100.1 vs encorafenib 8.13 / 29.18 / 105.5; HT29 mosperafenib 9.79 / 25.14 / 72.57 vs encorafenib 6.3 / 19.37 / 68.43. 'Encorafenib marginally more potent in both lines' holds at IC50 and IC80 but NOT at IC95 in LS411N (mosperafenib 100.1 vs encorafenib 105.5). Note also an internal inconsistency the audit missed: the S1 caption says 'IC50, IC90, and IC95' while the table reports IC80.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] LABEL ABSENCE RE-VERIFIED BY RETRIEVAL (audit's conclusion confirmed). openFDA drug label API returns NOT_FOUND for openfda.generic_name:"mosperafenib", for free-text 'mosperafenib', and for 'RG6344'. ClinicalTrials.gov API v2 returns zero studies for 'mosperafenib', 'RO7276389' and 'RG6344'. All re-run 1 Sep 2026. Label tier is empty — confirmed, not assumed.
[mosperafenib (RG6344 / RO7276389) — data cut 1 September 2026] EMA ABSENCE IS NOT ESTABLISHED. The audit reports an EMA endpoint 401 and treats the absence as settled. I could not retrieve a conclusive EMA result either (the medicine finder is JS-driven). This remains an unretrieved negative and must be labelled as such rather than as a searched absence.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] EVIDENCE BASE INCOMPLETE — the audit's central factual assertion is false. It states 'Crossref returns only the two AACR meeting abstracts' and 'There is zero peer-reviewed primary literature.' Crossref returns FOUR works. The audit missed two clinical conference abstracts, both retrieved and read in full: (1) ASCO 2025, Li J, Xu T, Zhang W, Li W, Kang Z, Yuan Z, et al. 'A phase 1, multicenter, open-label study of HSK42360, a brain-penetrant BRAF inhibitor, in patients with BRAF V600-mutated solid tumors.' J Clin Oncol 2025;43(16_suppl):3109. DOI 10.1200/jco.2025.43.16_suppl.3109; (2) SNO 2025, Li W, Kang Z, Shen L, Li S, Huang M, Zhang W, Mo L, Guo H, Zhang W. 'CTNI-28. Phase 1 study of brain-penetrant BRAF inhibitor HSK42360 in adult patients with BRAF V600-mutated advanced solid tumors or recurrent primary central nervous system tumors.' Neuro-Oncology 2025;27(Suppl):noaf201.0525. DOI 10.1093/neuonc/noaf201.0525. The audit's entire clinical section, its 'absence vs silence' cuSCC analysis, and its unknowns list were built on one third of the available clinical record.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] MOST IMPORTANT SUBSTANTIVE MISS — the sponsor's own primary-source mechanistic language is weaker than the audit recorded. ASCO 2025 #3109 and SNO 2025 CTNI-28 both state verbatim: HSK42360 'displays significantly less paradoxical activation than approved BRAF inhibitors and spares wt(BRAF)-containing RAF dimers.' AACR 2026 LB405 states 'with significantly reduced paradoxical activation compared to approved BRAF inhibitors.' The sponsor's claim in peer-facing venues is REDUCED, not ABOLISHED, paradoxical activation — a relative, comparative, unquantified claim. The 'spares wild-type-BRAF-containing RAF dimers' formulation is a distinct mechanistic claim that appears nowhere in the audit. This is the pivot of the whole paradox question and the audit did not record it.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] AACR 2026 LB405 MISCHARACTERISED. The audit states LB405 'is a clinical dose-escalation report... it contains no dimerisation experiment, no pERK or other pharmacodynamic biomarker data.' Retrieved abstract text refutes this: LB405 is a mixed preclinical + clinical abstract. Its Results section reports BRAF V600E glioma PDX tumour growth inhibition of 67%, 108% and 111.5% across three dose groups; brain distribution Kp,uu 1.09 (mouse) and 2.8 (rat); activity in a vemurafenib-plus-cetuximab-resistant colorectal cancer model where 'HSK42360 monotherapy outperformed the combination therapy'; and synergy with trametinib. The mechanistic sentence is presented as a summary of preclinical studies, not as a bare carry-forward assertion — though no method, assay, dose-response or figure is given in the abstract text.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] CLASS II — the audit's absolute negative is false. It states 'no class II mutant... is named or tested in any public preclinical or clinical material.' ASCO 2025 #3109 and SNO 2025 CTNI-28 both state: 'Treatment with HSK42360 results in excellent and durable anti-tumor effect in BRAF Class I and II mutant CDX or PDX models.' A conference-tier class II activity claim therefore exists and was missed. It remains evidence-free (no class II mutant named, no model named, no data shown) — but the correct finding is 'asserted at conference tier without data', not 'absent from the record'. CLASS III genuinely remains absent from every source; that part of the audit stands.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] FABRICATED FINDING — MUST BE WITHDRAWN. The audit's unknowns list asserts: 'The 22 June 2026 Nuvectis press release contains an apparent drafting error — it states the second-generation salt form of NXP200 was developed to enhance the PK profile of NXP100... Treat quantitative claims in that release with caution.' Retrieved verbatim text of the release: 'A second-generation salt form of NXP200 was recently developed to enhance the pharmacokinetic (PK) profile of NXP200, and early data indeed demonstrate a marked improved PK and greater single agent clinical activity.' There is no drafting error. The audit impeached the reliability of a primary company source on a misreading and then propagated that impeachment as a caution over all quantitative claims in it. Removed.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] TERRITORY MIS-SPECIFIED. The audit states rights were licensed 'outside Greater China.' The sponsor's own release text reads: 'Nuvectis in-licensed exclusive worldwide Ex-China rights for two drug candidates from Haisco. Haisco also retains rights for NXP100 in India and certain Southeast Asia territories.' 'Ex-China' is the primary wording; 'Greater China' appears only in Chinese trade/aggregator coverage (review tier). Whether China includes Hong Kong, Macau and Taiwan is not specified in any retrieved primary source. Additionally the India/Southeast Asia carve-out attaches to NXP100, not NXP200 — the audit did not record it at all.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] PHARMACODYNAMIC DATA — the audit's 'No pharmacodynamic biomarker data (tumour or skin pERK) at any dose, in any study' is false at the preclinical level. AACR 2024 #598 verbatim: 'Immunohistochemistry showed a close correlation between the inhibition of tumor growth and ERK1/2 phosphorylation in A375 CDX tumor tissues.' In vivo tumour pERK IHC exists. What is absent is CLINICAL PD (no tumour or skin pERK in any of the three clinical abstracts, despite pharmacodynamics being a stated secondary objective of NCT06536400 per ASCO 2025).
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] Kp,uu understated. The audit records only 'Kp,uu greater than 1' (AACR 2024). LB405 gives species-specific values: 1.09 in mice and 2.8 in rats.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] MEETING DATES WRONG. The audit dates AACR 2024 #598 to '22 March 2024' — that is the Crossref online-publication date. The citation format reads: 'American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA.' AACR 2026 LB405: '2026 Apr 17-22; San Diego, CA', Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts).
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] HGG RESPONSE QUALITY OMITTED. LB405 verbatim: 'In HGG, the ORR was 36.8% (7/19, including preliminary PR), and 1 patient had a confirmed CR.' The 36.8% figure includes unconfirmed responses. The audit reported 36.8% without this caveat, and reported '1 CR' without noting it is the only response described as confirmed.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] SELECTIVITY STATEMENT MISLOCATED. The audit places the unquantified wild-type selectivity claim at the enzymatic level. AACR 2024 #598 makes it at the CELLULAR level: HSK42360 'exhibited significant anti-proliferation activity against multiple tumor cell lines with BRAFV600E... but was highly selective for BRAF wild-type cells.' The 5 nM figure is enzymatic; the selectivity statement is cellular and separate. No fold-selectivity number is given in either place.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] EUROPE PMC RESULT MISSTATED. The audit says Europe PMC 'returns no primary research article.' It returns exactly one record: Cancer Innovation 2025, 'Re-Innovation in Clinical Trial Designs Based on Precision Therapy', DOI 10.1002/cai2.70028, pubType 'article-commentary' — a review-tier passing mention with no characterisation data. The peer-tier verdict (NONE EXISTS) is unchanged, but the search result was reported as empty when it was not.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] MISSED COMPARATIVE CLAIM. SNO 2025 CTNI-28 makes an explicit naive cross-trial efficacy comparison inside the abstract: 'An approved BRAF V600E inhibitor combined with a MEK inhibitor, had ORRs of 33% and 50% in BRAF V600E-mutated high-grade glioma (HGG) and low-grade glioma (LGG), respectively (dabrafenib US Prescribing Information, 2025).' This is the only label-tier document invoked anywhere in the NXP200 record — and it is a label for a DIFFERENT molecule, used as a historical benchmark. It establishes nothing about NXP200. The audit missed this claim entirely.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] MISSED COMPANY CLAIM. Nuvectis 22 June 2026 release, on the sodium salt: 'early data indeed demonstrate a marked improved PK and greater single agent clinical activity.' Uncited, no numbers, no denominators, no comparison shown. Company tier, asserted-only. Not in the audit.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] HARD-RULE VIOLATION IN THE AUDIT ITSELF. The binding_mode section reads 'Whether the molecule is a monomer-selective paradox breaker (plixorafenib/PLX-type mechanism) or a dimer-binding type II agent cannot be determined.' Plixorafenib is out of scope and must not be used as a mechanistic reference class here. Reworded without it.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] ABSENCE NOW VERIFIED BY RETRIEVAL, NOT BY FAILED SEARCH. Label tier: openFDA label API queried for both 'HSK42360' and 'NXP200' — NOT_FOUND for both; no approval exists in any jurisdiction; no US IND filed (expected 4Q2026); no NXP200-named study is registered on ClinicalTrials.gov. Structure: PubChem PUG-REST name lookup for HSK42360 and NXP200 — 'No CID found' for both; ChEMBL molecule search — 0 records. The audit's absence findings on label tier and chemical structure are CORRECT and are now retrieval-verified rather than assumed.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] TRIAL REGISTRY VERIFIED CORRECT. All four NCTs pulled from the ClinicalTrials.gov v2 API: NCT06536400 (Ph1, n=316, start 25 Jun 2024), NCT07158710 (Ph1 paediatric 6-<18y, n=159, start 15 Aug 2025), NCT07561554 (Ph1, HSK42360-Na, n=159, start 6 Mar 2026), NCT07790510 (Ph1/2, +cetuximab, mCRC, n=180, start 12 Aug 2026). All recruiting, all sponsored by Haisco, and all four eligibility criteria confirmed to require documented BRAF V600 mutation positivity. A registry-wide search for HSK42360 returns exactly these four and no others. The audit is correct here.
[NXP200 (HSK42360 / HSK42360-Na) — data cut 1 September 2026] ESMO DETAIL ADDED AND SOURCE-CORRECTED. Q2 2026 release verbatim: 'Oral presentation of additional clinical data generated by Haisco will be presented at the upcoming European Society for Medical Oncology (ESMO) conference (Madrid, Spain, October 23-27, 2026).' The audit gave no dates. Separately, the toxicity-reduction quote the audit attributes to release 3338215 could NOT be located in that release on retrieval — attribution flagged as unverified.
[tovorafenib (OJEMDA) — data cut 1 September 2026] VENUE INTEGRITY — MOST DAMAGING ERROR: the CHMP Assessment Report (EMA/67438/2026) is tagged venue="label" on two entries (the CHMP paradoxical-activation assessment, and the FIREFLY-1 Secondary Primary Malignancy AESI). Under the audit's own taxonomy "label" = an approved regulatory label (FDA USPI / EMA SmPC). A CHMP assessment report is a regulator-authored assessment document, NOT label text. Retagged to venue="regulatory-assessment". Consequence: the audit's headline finding — 'a regulator expressly contradicts the anti-paradox claim' — is real and regulator-authored, but it is NOT in either approved label. Neither the USPI nor the SmPC contains the word 'paradox'. Verified: grep -i paradox on the extracted SmPC returns 0 hits; DailyMed retrieval of the USPI confirms 0 hits. The audit's own citations block already hedged this as '[LABEL/REGULATORY]', inconsistent with the tag it used inside paradox_claims.
[tovorafenib (OJEMDA) — data cut 1 September 2026] FACTUAL, HIGH-STAKES: 'That is the entirety of US PI section 12.1' is FALSE. Section 12.1 has a second sentence. Verbatim (DailyMed SPL setid ea3a9631-3a66-6a7c-e053-2995a90ae2ad, rev. 8/2025): 'Tovorafenib is a Type II RAF kinase inhibitor of mutant BRAF V600E, wild-type BRAF, and wild-type CRAF kinases. Tovorafenib exhibited antitumor activity in cultured cells and xenograft tumor models harboring BRAF V600E and V600D mutations, and in a xenograft model harboring a BRAF fusion.' The word-absence finding itself (paradox / dimer / SCC / keratoacanthoma absent from the US label) is CONFIRMED on retrieval.
[tovorafenib (OJEMDA) — data cut 1 September 2026] CHMP did NOT reproduce the sponsor's ARAF value. EPAR p.32 verbatim: 'In vitro, tovorafenib inhibited the kinase activity of BRAF V600E, WT BRAF, and WT CRAF kinases, with IC50 values of 7.1, 10.1, and 0.7 nM, respectively. However, activity on ARAF was not presented, but it was reported in the literature that tovorafenib is not potent against ARAF (Tkacik et al., 2023).' The audit wrote that the four-value panel including ARAF 55 nM was 'reproduced by CHMP (EPAR p.33)'. CHMP reproduced three values, not four, and the ARAF 55 nM figure is Rastogi 2025 Table 2 only (verified independently).
[tovorafenib (OJEMDA) — data cut 1 September 2026] SELECTIVE REPORTING IN THE COOPERATIVITY DATASET: Tkacik 2023 reports Hill slopes ARAF-SSDD -1.12, BRAF-WT -2.89, BRAF V600E -0.75, CRAF-WT -1.55 ± 0.43, CRAF-SSDD -3.18. The audit omitted CRAF-WT (-1.55) and asserted a '-2.6 to -3.2 range for dimers' that does not exist in the paper. Only 2 of the 4 non-V600E constructs show steep slopes. The positive-cooperativity argument — the audit's stated 'strongest primary evidence underpinning the anti-paradox rationale' — is materially weaker than presented.
[tovorafenib (OJEMDA) — data cut 1 September 2026] CITATION ERROR: Kilburn et al. is Nature Medicine 2024;30(1):207-217 (published online 17 November 2023), NOT 'Nat Med. 2023;30(1):207-217'. Volume 30 issue 1 is the January 2024 issue. PMID 37978284 correct.
[tovorafenib (OJEMDA) — data cut 1 September 2026] FIREFLY-1 Arm 1 denominator is stated wrongly. Verified three distinct numbers: treated/safety N=77 (CHMP EPAR safety table, Arm 1; Nature Medicine); efficacy population N=76 (both approved labels); RANO-HGG response-evaluable N=69 (EU SmPC 5.1). The audit's 'n=76 treated / 69 response-evaluable' mislabels 76 as the treated population.
[tovorafenib (OJEMDA) — data cut 1 September 2026] EPAR page citations wrong: the Secondary Primary Malignancy AESI quote is on p.115/154, not 'p.~123'. 'Sensitivity to tovorafenib for Ras mutations and WT B-RAF was variable' is on p.32, not p.33. (The p.44 citation for the paradox assessment IS correct — verified verbatim.)
[tovorafenib (OJEMDA) — data cut 1 September 2026] NF1 GEMM ENDPOINT IS DESCRIBED THREE DIFFERENT WAYS ACROSS SOURCES and the audit reports only one. USPI 13.2 / EU SmPC 5.3: 'an increase in tumour volume was noted in 2/12 mice (approximately 17%)'. CHMP EPAR p.44: 'A subset of larger tumours (n=5) was also observed in treatment group with an increase in tumour NUMBER in 2/12 mice (approximately 17%)'. Rastogi 2025: 'A qualitative difference was observed in nine of 48 proximal nerve volume measurements in the tovorafenib-treated group which were relatively larger, (i.e., >3 mm3).' Volume vs number vs 9/48 measurements are not the same endpoint.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Rastogi synergy figure misreported: the audit gives 'Loewe synergy >=10', which is the significance THRESHOLD, not the result. Actual: Loewe 39.8 (Bliss 19.0) in sNF96.2; Loewe 27.2 (Bliss 29.8) in the ERMS PDX ex vivo. Combination partner pimasertib confirmed.
[tovorafenib (OJEMDA) — data cut 1 September 2026] US PI 12.2 is not exclusively exposure-response. It contains two subsections: 'Exposure Response Relationships' and 'Cardiac Electrophysiology'. The audit's 'contains only exposure-response statements' is incomplete. The load-bearing part of the finding — no MAPK pharmacodynamic claim in 12.2 — is CONFIRMED.
[tovorafenib (OJEMDA) — data cut 1 September 2026] IDENTITY: the December 2019 asset purchase was executed by DOT Therapeutics-1, Inc., a Day One subsidiary — not by Day One Biopharmaceuticals, Inc. directly. Sunesis Pharmaceuticals is now Viracta Therapeutics, Inc.; Takeda assigned its exclusive Sunesis/Viracta licence to DOT-1. Date 16 December 2019 confirmed (Day One 10-K).
[tovorafenib (OJEMDA) — data cut 1 September 2026] IDENTITY: the Ipsen ex-US licence date was missing. Executed and announced 25 July 2024 (~$111M upfront: ~$71M cash + $40M equity; up to ~$350M milestones; tiered double-digit royalties from mid-teens). Ipsen Pharma as EU MAH is confirmed on the EMA EPAR page.
[tovorafenib (OJEMDA) — data cut 1 September 2026] IDENTITY: alias list is correct as far as it goes (MLN2480, TAK-580, BIIB-024, DAY101, CAS 1096708-71-2 — all confirmed) but incomplete. Also carries AMG-2112819 and BSK1369; UNII ZN90E4027M.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Clayton et al. is no longer a preprint. It is eLife reviewed preprint 95334 v3 (14 January 2025) with a Version of Record; DOI 10.7554/eLife.95334.3. The audit's '[PEER, preprint stage]' tag is out of date. [FLAG: confirm VoR publication date — one retrieval reported 13 February 2025.]
[tovorafenib (OJEMDA) — data cut 1 September 2026] BROKEN CITATION: https://www.accessdata.fda.gov/drugsatfda_docs/label/2024/217700s001,218033s001lbl.pdf returns HTTP 404. The valid retrieval path is the DailyMed SPL (setid ea3a9631-3a66-6a7c-e053-2995a90ae2ad), version dated 2 September 2025, carrying 'Revised: 8/2025'.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Adelmann 2016 panel is not the three compounds the audit implies. Verified values are correct (vemurafenib 5.5, dabrafenib 10, encorafenib 50) and tovorafenib/MLN2480/TAK-580/DAY101 are confirmed ABSENT from the paper. However Table 1 contains a fourth compound which is out of scope for this audit and is not characterised here. The audit should not present the panel as complete at three.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Rasco 2023: 'keratoacanthoma 0/149' and 'no new primary melanoma reported' are stated as counted zeros. The paper reports the 1/149 cuSCC figure explicitly; the KA and new-melanoma figures are ABSENCE OF REPORT, not reported zeros. Restated as 'not reported'.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Rasco 2023 NRAS denominator: verified evaluable NRAS-mutant melanoma = 14 (Q2D treatment-naive) + 17 (QW treatment-naive) = 31, not 32. [FLAG: the audit's extra 'Q2D pre-treated n=1' NRAS patient was not located in the retrieved full text — the Q2D pre-treated expansion cohort is described as BRAF-mutant.]
[tovorafenib (OJEMDA) — data cut 1 September 2026] SNO 2025 three-year follow-up: the audit asserts the release stated 'no new safety signals' and lists specific adverse reactions. That wording was NOT found in the retrieved release, which discloses median study duration 40.6 months, 6 June 2025 cutoff, median retreatment duration 9 months, and median maximum tumour reduction -38.3%. The safety wording is unverified and has been removed.
[tovorafenib (OJEMDA) — data cut 1 September 2026] Sun 2017 corrigendum is reported one-sidedly. The corrigendum also states that the central conclusions remain valid and have been confirmed by subsequent research, and that the plastic-affinity issue does not affect clinical trials or animal studies using API. Added for balance. (Both the vinculin/Figure 5 A-and-C duplication and the borosilicate-coated glass tube omission are CONFIRMED verbatim.)
[tovorafenib (OJEMDA) — data cut 1 September 2026] Tkacik disclosure understated: Eck holds a Novartis consultancy AND sponsored research support from Novartis Institutes for Biomedical Research (plus SpringWorks). Novartis is naporafenib's sponsor and naporafenib is the head-to-head comparator in that paper. The paper remains non-sponsor with respect to tovorafenib, but 'independent' should carry this qualifier.