Most shelved assets were never cleanly falsified.
Was it the molecule, or the trial — and could a redesign bring it back?
A failed readout is usually reported as a failed compound. Often it is a failed question: a trial under-powered for the effect it was looking for, enrolling a population the mechanism could not help, or reading out on an endpoint the biology was never going to move. Revive separates those cases, because they have opposite consequences — one closes a programme, the other prices a second attempt.
A failed readout is not a clean verdict. Revive re-scans the graveyard and lifts the assets whose failure was the trial's design, not the molecule — and says plainly which ones stay down. A hand-authored schematic, not a measurement.
What Revive actually does for you.
A shelved asset is not free — it stands for years of work and hundreds of millions of dollars already spent. Yet most failed trials were never a clean verdict that the drug does not work. They were under-powered for the effect they sought, enrolled a population the mechanism could not help, or read out on an endpoint too blunt to register it — a failed question wearing the costume of a failed compound.
Revive performs the forensic autopsy. It separates the fixable trial-design mistakes from the molecule itself, and prices what a corrected trial would have to look like to bring the asset back — up to and including the honest verdict that it cannot.
Point Revive at a failed asset. It tells you whether the failure was the design or the drug, and what a redesign would need to constitute a real test — so you can price a shelved asset on evidence, not a hunch.
Counterfactual reconstruction over the one shared trial representation. It separates aleatoric molecule-level signal from design-induced variance, then estimates the corrected-design outcome interval with the same calibrated head and a signed audit row per verdict.
Was it the molecule, or the trial? The same failed readout, decomposed. Most of the shortfall is design — under-powering, the wrong population, an endpoint too blunt to move — and design is fixable; the molecule's own contribution is small. Corrected, the projected outcome clears the bar. A hand-authored synthetic fixture, not a measurement.
A write-off closes the file. Revive reopens it.
The default reaction to a failed Phase III is to shelve the asset — one binary verdict, no decomposition. But a null result and a real drug that was tested badly look identical from the outside. Telling them apart is the whole game, and it is out of reach for anyone who reads the readout as a verdict on the compound alone.
The inclusion rule wrote off a population the drug could treat.
It wasn't the biology. It was where the line was drawn.
For years, breast-cancer patients whose tumours were graded HER2-negative were treated as non-candidates for HER2-directed therapy — the trial inclusion rule drew a hard line at HER2-positive.
DESTINY-Breast04 tested a newer, more potent HER2-directed antibody-drug conjugate in the HER2-low patients on the wrong side of that line — and they benefited. The lesson is not that the old label was a clerical error; it is that a binary inclusion definition, paired with the right molecule, had been hiding a treatable population.
That is the question Revive asks in reverse: was the failure the compound — or a design choice, like an inclusion rule, that a corrected trial could fix?
Public record (Modi et al., NEJM 2022; DESTINY-Breast04, NCT03734029). Talaria was not run on this trial — it is shown to make the mechanism concrete. No Talaria prediction is attached to any real molecule.
Four things travel with every answer.
An attribution across the failure taxonomy
Efficacy, safety, enrollment, regulatory, competitive. Which mode actually claimed this trial, and how confidently that can be assigned given what was reported.
A falsification verdict
Whether the readout genuinely tested the mechanism, or whether the design made a null result close to inevitable regardless of the compound's merit.
The requirements a corrected trial would carry
Not a promise of success — the concrete design changes needed for a second attempt to constitute a real test, and the evidence that would have to exist first.
An explicit recoverability position
Including the cases where the honest answer is that the asset is dead and the evidence says so plainly.
Everything in a Revive dossier.
Revive is Built. Each verdict travels with the evidence and the caveats behind it — and, where the evidence supports it, the shape a second attempt would have to take.
Failure-mode attribution
Which taxonomy mode claimed the trial, and how confidently.
BuiltFalsification verdict
Did the readout test the mechanism, or was a null near-inevitable?
BuiltCorrected-trial requirements
The concrete design a real second test would need.
BuiltRecoverability position
Including the honest “dead” when the evidence says so.
BuiltComparable-precedent evidence
What has to exist first for a second attempt to be informative.
BuiltSigned audit row
A verifiable record behind every verdict.
BuiltRevive is Built. Where the evidence cannot support a claim, the output says so rather than inventing one.
One substrate, addressed at a different moment.
Revive runs the same substrate backwards. Where Intelligence asks what a draft would produce, Revive takes what a completed trial did produce and asks which design choices are sufficient to explain it. The taxonomy is the spine: a result consistent with several failure modes is reported as consistent with several, rather than resolved into whichever single story is most convenient.
The same substrate, run backwards. Where Intelligence asks what a draft would produce, Revive asks which design choices are sufficient to explain what a completed trial did produce. A hand-authored schematic, not a measurement.
Revive does not predict that a revived asset will succeed — it prices the conditions under which a second attempt would be informative. It is bounded by what was reported: trials that never posted results constrain what can be concluded, and that constraint is stated, not absorbed. Every figure shown publicly is a hand-authored synthetic composite — no sponsor, molecule or NCT beside any probability. No FDA authorization exists; pre-deployment, not for clinical use.
An interactive demo on a synthetic composite.
Every value in the demo is a hand-authored fixture. It is there to show the shape of the output and how it responds — not to report a measurement.