Fore Biotherapeutics · Regulatory Affairs · AI Support

Questions for agent training

Two things live here. Up top, the regulatory topics we’re grounding the agent on, organized by category. Below, eighteen items across three passes where your regulatory judgement is genuinely required — mostly confirmations rather than compositions, since we’ve pre-filled everything a public source can answer.

CONFIDENTIAL — internal. Precedent facts below are public-domain; Fore program specifics stay internal. Every output the agent produces is a draft requiring regulatory review — it is decision support, not regulatory or legal advice.
Training content · by category — tap a card to filter the questions below

Training topics

The compiled topics feeding the agent’s training — use the category cards above to filter.

Reference · read before answering

What we already hold

Everything below is drawn from public FDA/EMA records and the topic list you and Franklin compiled. Please correct what is wrong rather than restate what is right — anything here becomes a fixed constant the agent cannot override.

Facts — anchored regulatory precedents (public-domain)

Quoted or directly transcribed from FDA labels / approval summaries. If any is not how you want it framed, that is the most urgent thing on this page.

ItemWhat the record says
Tovorafenib (Ojemda)Accelerated approval 23 Apr 2024; patients ≥6 months; relapsed/refractory pLGG with BRAF fusion/rearrangement OR V600 mutation. ORR 51% (RAPNO-LGG, BICR, N=76); median DoR 13.8 mo. Type II pan-RAF inhibitor. Confirmatory FIREFLY-2 (NCT05566795) required.
Dabrafenib + trametinib (peds LGG)Traditional/full approval 16 Mar 2023; ≥1 yr; BRAF V600E LGG requiring systemic therapy. Randomized Ph2 (N=110): ORR 47% vs 11%, mPFS 20.1 vs 7.4 mo (HR 0.31). Type I inhibitor.
Dab + tram (tumor-agnostic)Accelerated approval 22 Jun 2022; ≥6 yr; unresectable/metastatic BRAF V600E solid tumors after prior treatment; excludes colorectal. Covers glioma.
Approval pathwaysTraditional/regular = confirmed clinical benefit. Accelerated = surrogate endpoint (ORR/DoR); confirmatory trial required and (per FDORA 2022) generally must be underway at approval.
Response criteriapLGG response assessed by RAPNO-LGG / RANO-LGG (T2/FLAIR-based), not RANO-HGG. Tovorafenib's FDA figure is RAPNO 51% — not the Nature Medicine RANO-HGG 67%.
US pediatric frameworkPREA obligation; iPSP/PSP agreed with FDA prior to NDA; RACE for Children Act (FDARA molecular-target list) drives pediatric oncology study requirements.
EU pediatric frameworkPIP agreed with the PDCO; class/condition waivers and deferrals; EMA conditional approval convertible to standard on confirmatory evidence.
Project ORBISFDA OCE framework for concurrent oncology review with international partners (Health Canada, TGA/Australia, Swissmedic, MHRA, ANVISA, HSA, others).
Post-marketingPMR = required study/trial; PMC = commitment. Follow accelerated approvals and many rare-tumor approvals to verify clinical benefit.

Assumptions — ours, and the ones most likely to be wrong

These are inferences shaping how the agent will operate. A single line of correction on any is worth more than a paragraph elsewhere.

We have assumedBecauseConfirm?
Grounding is public-domain only — Drugs@FDA, FDA guidances/approval summaries, EMA EPARs, PubMed, ClinicalTrials.gov, open webMatches our other tools; no licensed feed namedYes / No — add sources:
The lens is Fore's pediatric BRAF-CNS program (plixorafenib) as the anchorYour program focusYes / No
Every answer is a cited, HITL draft with verify-flags on anything unconfirmedRegulatory strategy must be defensibleYes / No
Topics split into on-demand precedent Q&A vs scheduled monitoring per the platform tagsSome topics are one-time deep dives, others recurringYes / No
Home is a standalone platform, later foldable into the FRED hubFastest path; reuses the agent pipelineYes / No
Tovorafenib's subgroup ORRs (fusion 52% / V600 50%) are exploratory, not poweredLabel lists them as exploratory; V600 n=12Yes / No
The breadth rationale is mechanism (type II pan-RAF), NOT the RACE Act/iPSPPrimary sources do not state RACE/iPSP as FDA's reasoningYes / No
Outputs are decision support, never regulatory/legal adviceHITL review chain owns the decisionYes / No
Known unknowns — we could not determine these
Pass 1 · about 15 minutes

Confirm, don’t compose

Ticks and corrections against material we have already assembled. Please correct rather than recall — a remembered figure that hardens into the system is very hard to dislodge later.

Facts and how they may be used

Each becomes a fixed rule the agent cannot override.

1

Using the Facts table above — the anchored precedents — mark any that are not framed the way you want, or that you would not want the agent to state without a caveat.

Especially the approval-type distinctions (accelerated vs traditional) and the tovorafenib breadth rationale.
2

The Assumptions table sets out grounding, lens, run-modes and the HITL bar. Correct anything wrong — particularly whether we may use only public-domain sources or you have licensed feeds to add.

This determines what the agent is allowed to cite and how current it can be.
3

Confirm the run-mode assignment for the 18 topics (on-demand precedent vs monitoring vs both), and give us the build priority order.

We will build in the order you set; the platform blueprint has our suggested tags to react to.
4

Below is the full set of rules we intend to write into the agent. Mark each keep / reword / not a rule, and add anything load-bearing we have missed.

Recognising a rule is far more reliable than recalling one cold — that is why this is a list, not an open question.
#Rule as currently written into the agentKeep / reword / not a rule
R1Approval type is stated precisely. Traditional/regular vs accelerated (surrogate endpoint, confirmatory trial pending). Never call an accelerated approval simply “approved” without the qualifier.
R2Every regulatory claim is cited to a primary source (FDA label/guidance/approval summary, EMA EPAR, peer-reviewed article). No uncited assertions.
R3Do not attribute a rationale to an agency unless a source states it. (E.g., RACE Act / iPSP is not FDA's stated basis for tovorafenib's breadth — mechanism is.)
R4Endpoints/criteria stated as used. Name the response criterion (RAPNO-LGG vs RANO-HGG) and never conflate figures across criteria.
R5Jurisdiction is explicit. FDA ≠ EMA ≠ MHRA/Health Canada/PMDA. Do not generalize one agency’s position to another without a source.
R6Separate “approved” from “precedent/guideline” from “proposed/under review.” Never present a precedent as a rule or a proposal as settled.
R7Flag exploratory / subgroup / small-N data as such (e.g., tovorafenib subgroup ORRs; V600 n=12).
R8If a precedent or figure is not in a locatable source, it is unknown. Say so and route to the research-gap list. Never estimate or fabricate.
R9Keep conditional/pending language conditional — “confirmatory trial ongoing,” “continued approval contingent” — never upgraded into settled fact.
R10Decision support, not advice. Every output is a HITL draft for the regulatory review chain; it does not constitute regulatory or legal advice.
R11Confidentiality boundary. Public precedents are public; Fore program specifics, strategy, and internal documents stay internal and are never echoed into a public-grounded answer.
R12End with the single most decision-relevant open flag — the next thing a human should verify.
+Anything load-bearing we have missed:

Precedent scope

The most common way a regulatory precedent gets misapplied.

5

For a precedent like tovorafenib’s broad accelerated approval, what makes it transferable to Fore’s program versus fact-specific — and where would leaning on it be indefensible?

The agent needs this to cite a precedent honestly rather than as a guarantee.
Pass 2 · about 20 minutes

Judgement only you have

Short prose. Two or three sentences each is genuinely enough.

Strategy & positioning

6

Which of the 18 topics map to a live or near-term Fore filing decision (iPSP, PIP, extrapolation, confirmatory-evidence design) — where the agent’s output would actually feed a submission?

7

What is the honest regulatory positioning of plixorafenib’s pediatric BRAF-CNS path — the argument you would defend in a Type B/C meeting?

8

Where is cross-jurisdiction extrapolation (FDA↔EMA↔MHRA/HC) legitimate, and where does it become misleading?

A verbatim caution sentence is ideal — we will hard-code it.

Boundaries

9

What must the agent never do, even when a user pushes — e.g., predict an approval outcome, assert an agency will accept a design, or offer legal conclusions?

10

What language reliably signals over-reach in a regulatory answer? Specific words or constructions you flinch at (“FDA will require…,” “guaranteed pathway,” “equivalent to approval”).

11

Name three sentences — actual sentences, not categories — that would make you wince if the agent produced them.

These become adversarial test cases. Concrete beats abstract here.
12

How current must an answer be to be trustworthy — and what should the agent say when the most recent guidance/approval may post-date its sources?

A verbatim “verify against the current record” holding sentence would help.

Comparators & landscape

13

For the BRAF/MAPK competitive and emerging-inhibitor monitoring, which competitors and assets must never be missed — and do our proposed keywords (exarafenib, NST-628, PF-07799933/RG6344, ABM-1310, ERK inhibitors, next-gen pan-RAF) match your intent?

14

What does a useful monthly brief look like to you — what would you actually read, and what is noise?

The room

15

Give us five regulatory questions you expect to be asked (internally, or by a partner/board) about the pediatric BRAF-CNS path, with the answer you would want given.

These load straight into the agent’s question-bank and double as evaluation cases.
16

Where is the line between a defensible precedent summary and a strategic recommendation that needs your sign-off? That boundary is what we most need to encode.

Pass 3 · optional, any time

The calibration set

The highest-value thing you could give us, and the least urgent. It can wait weeks without holding anything up.

17

What is the single regulatory question you are asked most often — and roughly how do you answer it?

Two or three sentences. This one alone is worth having.
18

If you have half an hour at any point, four or five more real questions with your own answers would give us a proper calibration set.

Forwarding an existing email thread or advisory Q&A works as well as writing new ones — we will extract the pairs. Please exclude anything privileged or partner/agency-confidential.

Add a question

Internal working copy — anyone reviewing this can add a topic or question. Submissions appear below and feed the next revision.

Added topics appear in Training topics near the top of the page.

💬 Add Comment
💬 Discussion