Stop reading the panel. Start moving the levers.
Drugs that succeed in Phase II routinely fail in Phase III. The failure is rarely the molecule — it is the translation: the endpoint, the population, the powering, the enrichment cutoff. Below is a bench. Move a lever and watch the shape of the answer change.
Everything on this bench is a hand-authored fixture. It is not the Talaria engine, and it is not about any real drug. We print its entire arithmetic further down the page, because a bench you cannot audit is a bench you should not believe.
The lever this studio lets you move is the one that flipped a real drug's fate.
Gefitinib in a broad, unselected lung-cancer population (ISEL, vs placebo) showed no clear benefit. The same drug, tested first-line in patients selected for an EGFR mutation (IPASS), beat chemotherapy on the progression-free endpoint in mutation-positive patients — and did worse in mutation-negative ones. The molecule never changed; the selection decision did. Below, move that lever yourself and watch the projected interval cross the bar.
Public record (Thatcher et al., Lancet 2005; Mok et al., NEJM 2009). Talaria was not run on this trial — it is shown to make the mechanism concrete. Every value in the interactive demo below is a hand-authored synthetic fixture, not a measurement.
Move one thing
In plain terms: move a single design lever and watch the calibrated outcome interval shift relative to the clearance threshold — the same kind of decision that separated ISEL from IPASS above. Every value below is a hand-authored synthetic fixture, not a measurement.
The expression distribution in this field is arbitrary. It is a fixture, not a prevalence estimate, and the scale is unitless — it is not TPS, not a percent, and not any real assay.
The line across the top of the field is the fixture curve for lever 01 under the current bundle — hand-authored, drawn, and named. It is not a sensitivity analysis.
A hand-authored fixture built to show the shape of the answer. Not a Talaria prediction. Not about any real drug.
The field renders the population. Lever 01 is the lever that changes the population, so lever 01 is the lever the field shows. That is a rendering choice, not a ranking.
The faint bar is this run's Day-0 shape; the solid bar is the shape under your bundle. A display convention for a fixture, not a result. Renormalizes to 100% of the residual failure shape. Enrichment is not free. It buys signal with accrual.
What a counterfactual panel looks like
The bundle diff: what you moved, and away from what. One joint schematic outcome for the whole bundle — never a per-lever delta, because a per-lever delta is a ranking.
| Lever | Day-0 | Now | Causal interval · sensitivity |
|---|
The fourth column is empty for all five levers, including lever 01. In production each lever carries its own causal interval and sensitivity analysis. This page carries none, for any lever, because this page has no engine behind it.
Joint schematic outcome for this bundle: 0.41 (schematic)
Two trials, one class, opposite outcomes
| Population as designed | Outcome |
|---|---|
| Broad, all-comers | MISSED primary endpoint |
| Above a high PD-L1 expression cutoff | HIT primary endpoint |
In first-line non-small-cell lung cancer, two checkpoint-inhibitor Phase III trials read out within a year of each other. One enrolled a broad, all-comers population. The other enrolled only patients above a high PD-L1 expression cutoff. The broad trial missed its primary endpoint. The enriched trial hit. The molecule was not the difference. The enrichment cutoff was the difference — a single design parameter, chosen at design time, free until the moment it wasn't.
This register holds facts and has no slot for a probability. It names no sponsor, no molecule and no NCT ID, and it never receives a number from the bench above.
The whole fixture, printed
Every number this page has shown you came out of the constants below. They are hand-authored. We publish them in full, at body size, because the alternative is asking you to trust a bench.
The real engine ships with its levers attached
This bench is a fixture. The engine behind M01 is not — and every prediction it emits arrives with a counterfactual panel, an uncertainty envelope, and an audit trail. Talaria is pre-deployment. We are opening a founding cohort ahead of it.
Join the founding cohort