Fore Biotherapeutics · Finance & IT · plixorafenib (PLX120-03 / FORTE)

Regulatory Evidence Platform — CFO Summary

Subject: Supporting Regulatory capacity through IT + the AI platform, without added headcount Prepared: 11 Aug 2026 Classification: Internal — Confidential

Bottom line

In a single working session, the platform produced the complete disease-background evidence base for CTD Module 2.5.1 across five CNS histologies — six linked documents, 65 distinct sources verified to their exact location in the primary literature, and a traceability register mapping every claim to its source.

It also found eight material errors in our own working material — including a competitive benchmark we were overstating and a Breakthrough Therapy designation missing from the development rationale entirely.

The financial argument is not headcount substitution. It is that the expensive failure mode in regulatory work — an unsupported number reaching a submission, an investor deck, or a reviewer — is now caught before it leaves the building, by a process that runs in hours rather than weeks and leaves an auditable trail.

1 · What was delivered

Documents

1 CTD 2.5.1 rationale · 1 traceability register · 4 disease references

Histologies

DLGNT · PXA · pilocytic astrocytoma · GBM · epithelioid GBM

Sources verified

Each pinned to its table, figure or named section

Material corrections

Found in our own working material

Open items surfaced

10 evidence gaps + 2 strategic decisions, each with a named owner action

Working session

Draft-to-deployed, reviewable by the team the same day

The dashboard is the single entry point — status, coverage, open items and every source, in one place, reviewable by anyone with the link.

2 · What it caught — the part with money attached

These are not typos. Each is a claim that had already been written down and would have travelled into downstream material unless someone re-read the primary sources.

Finding	What was wrong	Why it matters commercially
Competitive benchmark overstated	Our working material cited the competitor's efficacy as ORR 67% / DOR 16.6 months. The approval figures are ORR 51% / DOR 13.8 months — a different cut of the same trial.	The bar we benchmark against is checkable against the competitor's label. Quoting the higher pair in an investor or partnering conversation invites a correction from the other side of the table.
Our own strongest regulatory asset was missing	The development rationale did not mention plixorafenib's Breakthrough Therapy designation (1 Apr 2026, adult BRAF-V600E high-grade glioma) — reported as the first BTD for a targeted therapy in HGG. Fast Track and Orphan Drug were also absent.	This is the single most valuable regulatory fact we hold. Omitting it understates the asset in exactly the documents where it should be doing the most work.
The relevant approval precedent was absent	The rationale cited only one precedent. It missed the tissue-agnostic BRAF-V600E approval (22 Jun 2022) — the structural template for a molecularly-defined programme like FORTE.	Precedent is leverage. An approval pathway that has already been granted to someone else is the cheapest argument available to us.
A statistic quoted without its significance test	A trial comparison was carried as "39% vs 52%" with no mention that P = 0.10 — the difference was not statistically significant.	Presented that way it implies a demonstrated advantage the trial did not show. A reviewer or a diligence analyst checks this in one search.
Inconsistent numbers across our own readouts	Our published CNS response rates appear with three different denominators — and two different "67%" figures — across separate readouts.	Two Fore documents quoting different numbers for the same thing is a credibility event, in diligence or in an agency interaction. Now flagged and controlled.
A supporting study that does not support the claim	The reference used to link BRAF to glioblastoma states plainly that it contains no epithelioid GBM cases, has five BRAF-mutant tumours (one canonical) and bases a survival claim on three patients.	Building a rationale on it would not survive scrutiny. It has been formally retired from the argument with the reason recorded.
Two studies treated as independent that are not	Two papers presented as independently disagreeing share their senior authors, and up to ~12 of 30 patients may overlap.	Double-counting evidence overstates certainty in both directions. The caveat is now on the face of the claim.
An outlier figure that would have become "the number"	One single-institution series reports a mutation frequency of 100% where the literature says ~50%. It is an ascertainment artefact.	Marked "do not quote." A 100% figure is exactly the kind of number that gets lifted into a slide and then has to be walked back.

The pattern worth noticing. Six of the eight are errors of omission or over-claim in our favour — a benchmark too flattering to the competitor, a designation we forgot we had, a precedent unused, a significance test dropped. These are precisely the errors that internal review is worst at catching, because nothing in the document looks wrong.

3 · The operating model — why this does not require hiring

Regulatory evidence work has three components. They have very different cost profiles, and only one of them genuinely requires a regulatory professional.

Component	Who does it now	Cost behaviour	Effect of the platform
Source retrieval & verification find it, read it, pin the number to its table or figure	Platform	Was: the dominant time cost, and the first thing dropped under deadline	Effectively eliminated as a labour cost. 65 sources retrieved and located in one session. This is the bulk of the hours in a literature-grounded document.
Drafting & structuring assemble to CTD format, keep it consistent	Platform	Was: the usual reason to hire a medical writer or engage an agency	Collapsed to hours. Format compliance (ICH M4E) is a rule set — the platform applies it consistently across every document without drift.
Judgement & accountability is this the right argument · is this claim defensible · sign-off	Regulatory (human)	Unchanged — and should be	Amplified, not replaced. Reviewers now spend their time on the 12 flagged decisions rather than on assembling and re-checking the base material.

The honest version of the claim. This is not an autonomous system and should not be sold internally as one. During this build the platform mis-scoped the task twice, initially missed a whole section of the source corpus, and placed a link on the wrong page — each caught and corrected by the reviewer in the loop. That is the model working as designed. The value is not "no humans": it is that a small regulatory team can now produce and defend a volume of evidence work that would otherwise require either new headcount or an agency, while keeping sign-off exactly where it belongs.

3.1 · What IT actually had to provide

No new systems and no new licences beyond the AI platform already in place.
Hosting is the existing internal site — the six documents deploy to the same place the rest of the department's material already lives.
Review runs through the comment feature already built into every page: SMEs annotate inline, in context, and their corrections were incorporated the same day. Three named SME corrections are already recorded against these documents.
Output is plain HTML — no proprietary format, no vendor lock-in, and it survives independently of the tool that produced it.

4 · Cost frame

Read this before the table. No Fore-specific cost data was supplied to me — no salary bands, no agency rates, no budgeted headcount. The figures below are illustrative placeholders to show the arithmetic, not estimates of Fore's actual costs. Replace every input marked "assumption" with Finance's real numbers before this is used for any decision. The output line is only as good as those inputs.

Line	Input	Basis
Documents produced	6	Actual — verifiable on the dashboard
Distinct sources verified	65	Actual — each located to table/figure/section
Elapsed time, draft to deployed	1 working session	Actual
Conventional effort for equivalent scope	assumption	Medical-writing + literature-verification hours for a CTD 2.5.1 evidence base across 5 indications. Finance/Regulatory to supply.
Blended rate (FTE or agency)	assumption	Loaded FTE cost or vendor day-rate. Finance to supply.
Avoided incremental headcount	assumption	Whether a hire was actually planned for this scope. If none was planned, the saving is capacity, not cash — state it that way.
Rework / correction cost avoided	assumption	Cost of an unsupported claim reaching a submission, deck or diligence set. Low frequency, high severity — model as risk, not as run-rate.

How I would frame it to the board, given the above. Lead with capacity and risk, not with a savings number. The defensible claim is: "the same regulatory headcount now produces a fully-sourced, audit-traceable evidence base across five indications, and the review step is catching benchmark and precedent errors before they leave the company." A cash-savings figure is only credible once Finance supplies the rates — and it is the weaker argument regardless.

5 · What would undermine this case

Risk	Severity	Mitigation in place
Treating output as sign-off ready	High	Every document carries an explicit "medical + regulatory HITL review required" banner and is marked working draft. None is submission-ready and none claims to be.
Unverified items leaking into external use	High	Three items are flagged blocking for external use — conference abstracts and agency pages that blocked automated retrieval and need a manual pull. They are named, not buried.
Embargo breach	High	The embargoed CNS reinterpretation is excluded from all six documents by design, with the exclusion stated in each.
Confident-but-wrong output	Medium	The traceability register exists precisely so any claim can be checked back to source in one step. Contested figures are shown with both values rather than reconciled to one.
Key-person dependency	Medium	Output is plain HTML on the existing site with sources listed in full — readable and maintainable without the platform.
Scope creep into judgement work	Medium	The two strategic decisions surfaced were explicitly escalated as programme calls, not resolved by the platform.

6 · Recommendation

Do not budget a hire for this scope yet. The capacity constraint that would have justified it has moved — the binding constraint is now SME review time, not drafting or literature capacity.
Fund the review step instead. The value depends on named SMEs reading and correcting. That is a claim on existing people's calendars, and it is the thing to protect.
Close the three blocking items before any of this material is used externally — they need manual retrieval, roughly a day of a person's time.
Take the two strategic decisions to the programme: the adult designation against a paediatric positioning, and where we claim our space while the competitor moves front-line. Neither is an IT or writing question.
Ask Finance for the rate inputs in §4 if a cash figure is wanted for the board. Until then, present this as capacity and risk control.

Basis & limits. All counts in §1 and §2 are machine-derived from the six documents and independently checkable on the dashboard. The eight findings in §2 are recorded in the citation register with the corrected figure, the source, and the reason. §4 contains no Fore financial data — the rate, effort and headcount lines are unfilled assumptions and are marked as such; nothing in this document should be read as an estimate of Fore's costs or savings. The self-assessment in §3 is deliberate: the platform's errors during this build are stated because a business case that omits them would not survive its first challenge.