Asclepius
00-product-thesis.md

Product thesis — what Asclepius is, and what it's a proof of

Asclepius is a biotech research / analysis / advising tool. Its value is the creativity, architecture, and methodology it embodies — a framework that forms a defensible view on a single asset, shows its work, and names what would change its mind. Read this first; it frames what the rest of the methodology folder is for.

What Asclepius is

A biotech research / analysis / advising tool. You give it a pre-approval asset — sponsor, phase, mechanism, capital position, supply situation — and it produces a defensible view: a probability-of-success chain with a full audit trail, a phase-gated rNPV with a Monte Carlo band, a routed comparable cohort, and an investment memo that makes a call, states its conviction, and names the single readout that would flip it.

It is used by its author and a small trusted circle to analyze real frontier assets. It is not a public product, not a SaaS, and not a forecasting service.

What it's a proof of

The get-hired value of this project is not a track record. A live forecasting record takes years to mature and resolves after any plausible hiring decision. The value is the creativity, structure, and methodology the tool demonstrates — the kind of thing a skeptical biotech investor can read in an afternoon and conclude "this person thinks like an investor, and can build."

Three things carry that weight:

  1. Two named structural path-dependencies. Most valuation tools price a drug as a probability-weighted cash-flow stream and stop. This one treats two structural facts as first-class value drivers: who holds the capital (reflexivity — a Spence-style signaling multiplier on PoS keyed to sponsor capital state) and who controls the supply (supply constraint — a manufacturability multiplier on PoS and a peak-sales ceiling in rNPV). Both are reasoned structural hypotheses, grounded in named sources and honestly labeled as estimates, not calibrated corrections. They are the framework's point of view.

  2. A recommendation that earns itself. The memo synthesis doesn't print a number — it forms a view: an explicit conviction level (orthogonal to the buy/hold direction), a one-line verdict on how the path-dependencies net-moved the call, and a falsifiable kill criterion (a named event with a threshold and a direction). It shows its work and it says what would change its mind.

  3. An architecture that generalizes. A single DiligenceRecord is the source of truth; pluggable data-sources, analysis modules, and agents compose against it. Adding the supply-constraint dimension (a whole new path-dependency, with effects on two engines) required no core edits — it slotted into the same waterfall position reflexivity occupies. Every multiplier renders its real citation, including the frontier-modality ones that honestly say "Asclepius estimate, no BIO/Informa cohort." The audit trail is an engine output, not a UI decoration.

Methodology applied to its own surface

The design discipline runs one level deeper than the math. The diligence page is organized by the sequence of questions a senior investor actually asks (recommendation first, then thesis, valuation, operational diligence, and — co-equal — risk & limits), mirroring equity-research and IC-memo convention rather than the order the backend modules happen to register. And it is persona-aware: a VC associate, an IC voter, a scientific reviewer, and a quant skeptic each open the same record and get a different first screen, because they ask different first questions. The framework's methodology informs not just which numbers to compute but how to present them — the persona system is the methodology admitting its readers are not interchangeable. (Implementation: web/lib/persona-config.ts; design notes: docs/ia-redesign-notes.md.)

What it deliberately is not

  • Not a forecasting platform or a track record. The framework's value is its reasoning structure, not a scoreboard. A second-opinion ML PoS prior is included (a model that disagrees with the rule-based estimate is a signal), and a per-asset segment reference gives directional cohort context — both explicitly caveated as analysis, not accountability.
  • Not a public launch. It is gated for a small circle; there is no signup, billing, or growth surface.
  • Not calibrated truth. The path-dependency magnitudes are structural hypotheses, honestly labeled. The tool's claim is "here is a defensible, auditable way to think about this asset," not "here is the right answer."

See also