The Front Door · Calibration Sprint

AI-Oncology Calibration Sprint

A short, senior-led engagement that tells you — with reasons — whether your AI-oncology program, platform claim, or regulatory story is calibrated enough to carry into a serious FDA-facing or board-facing conversation, and whether the right move is to fit established precedent or to propose a reasoned departure the evidence can defend.

The scarce input is not the AI. It is the senior clinical-regulatory operator who knows when the AI is wrong — and has thirty years of FDA interactions to prove it. The Sprint puts that judgment on your specific decision, on a two-week clock, for a fixed fee.

Why This Exists

In AI-oncology, calibration failure takes two forms — and both are expensive.

The obvious failure is overclaiming. A program reaches the pre-IND or first-in-human stage carrying an endpoint the agency does not regard as a development endpoint, a trial design that cannot enroll its intended population, or a model whose provenance cannot be audited to regulatory standards. The story is ahead of the evidence.

The subtler — and often costlier — failure is premature conformity. A program forces itself into established precedent when the science, the mechanism, the biomarker, or the patient population may warrant a reasoned departure. Some of the most consequential oncology approvals — accelerated approvals on early response data, tumor-agnostic labels, novel surrogate endpoints validated under unmet need — came from sponsors who engaged the agency in territory the agency itself acknowledged was less well understood, and who built the evidentiary package that made the new path defensible.

The failure is not choosing one or the other — it is not knowing which case you are in, and not having the evidentiary package that would make the chosen path defensible. Both failures are failures of calibration between what the model or platform produces and what regulators have historically required — or would entertain. Most companies discover the misalignment late — inside a pre-IND meeting, or in front of a board — after capital and calendar time are already committed. The Calibration Sprint is designed to surface it before that.

The Question the Sprint Retires

Before we spend the capital and the calendar time — is our development story calibrated enough to survive the room it's about to enter? And are we forcing ourselves into precedent when the science warrants a reasoned departure — or proposing beyond precedent when the evidentiary package is not yet defensible?

You walk out knowing what is true, what is conditionally true, what is asserted but unsupported, and which of your next decisions carry the highest regulatory consequence — each with the reasoning trail, the strongest counterevidence, and the conditions under which the recommendation would change. And for the decisions where it matters, you walk out knowing which posture the evidence actually supports — fit the guidance, or propose a reasoned departure — and what it would take to make that choice defensible.

Who This Is For

A real decision your program turns on, in the next 90 to 180 days.

• Biotech-shaped AI-oncology companies running their own program toward an IND, with a real pre-IND or Phase I decision immediately in front of them.
• AI-platform companies selling into pharma — target discovery, patient stratification, dose optimization, biomarker development — whose commercial momentum depends on partners defending the platform's role to the FDA.
• AI-native drug developers selecting, designing, acquiring, or licensing their own oncology programs — where an independent clinical-regulatory view must test the resulting forecast, asset decision, trial scenario, or investment thesis.
• Boards, investors, and family offices — including philanthropic or venture-philanthropic oncology capital — evaluating an early oncology program, an indication choice, or a regulatory story before committing to the next round, a partnership, or a disciplined no.

You are a fit if the decision in front of you is real and your program turns on it — endpoint strategy, trial design, indication choice, or "can our partner defend this to a CDER reviewer?" — and you want senior judgment on it now, without first committing to a multi-quarter relationship.

Two variants, same arc, different interface

Biotech-shaped — the work happens at the agency–sponsor interface: target, indication, endpoint, population, trial-design and regulatory friction on your program. Platform-shaped — the work happens at the partner–platform interface: what regulatory-readiness support the platform can credibly offer partners, what must remain partner-owned, and which commercial claims will not survive a partner's regulatory team.

The Sprint converts into a longer arc where warranted — see the first-90-days notes for the biotech-shaped and platform-shaped variants.

Pre-IND Readiness Check

Where the decision in front of the program is whether to request a pre-IND meeting at all, a Sprint can be scoped around a readiness check. It retires one question:

Is this program ready to request the interaction — or would requesting now spend it on questions the team has not yet reconciled internally?

The check assesses readiness across six integrated domains, including the evidence that did not work — the rejected models, negative findings, and unvalidated assumptions a readiness position built only from supporting evidence leaves out. It returns a position for each domain, the places where two functions hold assumptions that cannot both be true, and a go-or-hold recommendation with the conditions that would change it. Where the answer is hold, it names the shortest path to ready.

It can return not ready. That is a useful answer, and it is cheaper than the alternative. Scoped within a Sprint or a longer engagement; the diagnostic itself is not published. Readiness is a statement about your program, not a prediction of the Agency's response. The reasoning behind it →

Independent oncology calibration for AI-native drug developers

When an AI-native developer selects or designs its own programs, the question is not whether its systems can generate a probability or thousands of scenarios. The question is whether the resulting oncology decision survives independent clinical, regulatory, enrollment, and board scrutiny.

Questions a Sprint may be able to retire:

• Does an oncology forecast remain calibrated when tested against evidence available before the outcome was known?
• Should a proposed oncology acquisition advance, and what evidence would change that conclusion?
• Do the proposed endpoint, population, indication, and trial scenarios survive oncology-specific clinical and regulatory review?
• Is the oncology investment thesis defensible to the board or investment committee, including the strongest countercase and a disciplined no-go path?

These are examples of questions OncAdios may scope as a Calibration Sprint, not claims of prior delivery or validated forecasting performance.

Inputs Required

• A specific, answerable question. The Sprint is accepted only when the question is sharp enough to retire in two weeks.
• Public-domain evidence, plus your own program materials — development plan, platform and target description, preclinical data, prior public data, the deck you would show an investor. Most of that is confidential, and it is expected to be: a mutual NDA is signed before any confidential material changes hands.
• Four categories are declined outright, and this is not negotiable: patient-level data or PHI; attorney–client privileged material; pre-filing draft patent claims; executed contracts or financial-account data. If any of it arrives, work stops until it is withdrawn.

How Your Material Is Handled

Handling is governed by the AI Agent Operating Charter, which routes every input by sensitivity rather than by convenience.

• Public-tier work happens in ordinary tools. Confidential client work happens inside a protected channel — a Tailscale-only, commercial-API-on-a-controlled-VPS architecture — and never in consumer AI tools.
• Where the analysis can run on de-identified or code-substituted material, it does. Naming discipline is the default, not a request: protected molecule descriptors never enter a deliverable that may leave the protected channel without explicit declassification review.
• The no-reconstructable-disclosure rule governs every citation and paraphrase: combined with publicly available information, a deliverable must not permit reconstruction of identifying client material.

The full operating standard →

Conflicts

OncAdios takes a small number of engagements in a narrow field. Before a Sprint is accepted, prior exposure to another program is checked for conflict. Where one exists the Sprint is declined and the reason given — that protects the company already engaged and the one asking.

What the Sprint Is Not

• Not a regulatory submission, and not representation before any agency.
• Not clinical-trial operations, legal, or financial services.
• Not a guarantee of FDA acceptance — no honest operator can deliver one.
• Not a name-on-deck retainer. It is a terminal, fixed-scope engagement that stands on its own.
• Not software. AI is the internal leverage layer under senior judgment and explicit governance. You are paying for the judgment, not the tool.
• Not a report factory. Limited availability; accepted only when the question is specific enough to answer well.

How It Connects

A complete engagement in itself — and the on-ramp to the deeper structures.

You can take the Calibration Map, Evidence Brief, and Board One-Pager and walk. Where it warrants, the Sprint is also the natural first rung:

Calibration Sprint → 90-day operating arc → Fractional CMO, Board or Scientific Advisor, or Co-founder.

If the Sprint surfaces that the real need is the next 8–12 decisions across a full pre-IND arc, the conversion path is explicit — and, where warranted, a portion of the Sprint fee may be credited toward a subsequent 90-day engagement by agreement. If it surfaces that the program needs less than you thought, the Sprint says so. Refusal is part of the output.

There is a second front door for a different question. Where the decision is not "will this story survive the FDA room?" but "what has this first-in-human program actually learned, and is the next dollar aimed correctly?", the FIH Diligence Read is the instrument — same fixed scope, same ten business days, built for investors, founders and boards at or after first-in-human. The fit call sorts which one you need.

See the full engagement structures and the operating standard behind every deliverable.