Pillar 6 of 6 · Verdict
Go, or no-go?
Everything above composed into one recommendation — with the risks that could still change it, and an honest account of how confident the answer actually is.
Go, or no-go?
What it answers
Go, or no-go?
The verdict is the point of the whole assessment, and it is also where most tools quietly overclaim. Here the gate is plain, unit-tested code, the probability of success is labelled directional everywhere it appears, and the combined confidence is the weakest leg rather than a flattering average.
What is the recommendation?
A single structured go/no-go composed from the probability prior, the competitive recruiting density and the cost band — decided in code, narrated afterwards, never decided by a language model.
How likely is this to succeed?
A directional probability of success: an empirical base rate for the phase and indication, adjusted by your trial’s own features, with each feature’s contribution shown as an attributable step.
What could still change it?
The operational risks the assessment can see — recruitment, cost, timeline, investigator bandwidth, site competition, safety signal — plus what the assessment thinks your study is overstating or has not yet considered.
Inside the pillar
6 analyses, each one named and cited
These are the analyses that actually run — not a category list. Each states what it computes and what it needs, because an analysis that quietly degrades is worse than one that reports it could not run.
Go / No-Go recommendation
Composes the probability-of-success prior, the competitive recruiting density and the per-patient and total cost band into one structured recommendation. The gate is plain, unit-tested code so that the call is auditable and reproducible; the language model only narrates the result. Combined confidence is the weakest leg, not the average.
Needs
The upstream pillar outputs. A missing leg lowers the confidence and is named, rather than being silently dropped so the verdict can still look decisive.
Probability of success
Starts from an empirical base rate for the phase and indication drawn from a historical trial-outcome ledger, then adjusts it in log-odds by standardized trial features assembled across the pillars, showing each feature’s contribution as an attributable step in a waterfall.
Needs
Phase, indication and the design features from the other pillars. The read is directional — a prior informed by public precedent and adjusted by your trial, labelled as directional wherever it appears. It is not a prediction and not a guarantee.
Operational-risk dashboard
A pure roll-up that reduces the other analyses to six traffic-light cells — recruitment shortfall, cost overrun, timeline slip, investigator bandwidth, site competition and adverse-event seriousness — each gated on whether its input analysis actually produced evidence.
Needs
The other analyses. It makes no new external call, and a cell whose input did not run stays unlit rather than turning green on absence.
Adverse-event signal
Runs a drug-versus-all-drugs disproportionality analysis against the pharmacovigilance background — proportional reporting ratio, reporting odds ratio with a confidence interval, and an empirical-Bayes shrunk score — rather than reporting a drug’s own top-reaction share as though it were a signal.
Needs
A named molecule, or its class for a pre-launch asset. Sponsor safety data from uploaded clinical study reports can be folded in on top of the public signal.
Sponsor delay signals (SEC filings)
Scans SEC filings and earnings-call transcripts for the sponsor’s own delay language and reduces the matches to a per-quarter ribbon, each flag carrying its filing date and the excerpt that triggered it.
Needs
A sponsor with public filings. A private sponsor files nothing to scan, and the analysis reports that rather than an implied clean record.
Verdict interpretation
Hands a compact, factual brief of every analysis snapshot to a language model under a strict guardrail and returns a structured critique: what the study does not yet consider, and what it overstates. The model interprets only the brief — no outside knowledge, no new numbers.
Needs
A completed assessment. Every claim in the critique is traceable back to a snapshot in the brief it was given.
Where the evidence comes from
Public sources, named
This pillar makes very few external calls of its own — most of it is a roll-up of the other five. That is deliberate: the verdict must be reproducible from the snapshots it composed, and a reader must be able to walk from the recommendation back to the individual analysis that moved it.
FDA
U.S. Food & Drug Administration · Weekly
Approvals, labels, guidance, and adverse-event signals (openFDA) used to anchor regulatory-uplift and safety risk.
open.fda.gov (opens in a new tab)ClinicalTrials.gov
U.S. National Library of Medicine · Daily
Registry of public and private clinical studies worldwide — enrollment, status, sponsors, sites, and outcomes.
clinicaltrials.gov (opens in a new tab)AACT
Clinical Trials Transformation Initiative (CTTI) · Daily
Relational, queryable database of all ClinicalTrials.gov content — the analytical backbone for site and enrollment statistics.
aact.ctti-clinicaltrials.org (opens in a new tab)PubMed
NCBI / U.S. National Library of Medicine · Daily
Biomedical literature index of 36M+ citations — used to ground indication baselines and standard-of-care evidence.
pubmed.ncbi.nlm.nih.gov (opens in a new tab)Also read by this pillar
What you get
What lands in the assessment
The recommendation names the analyses it composed and the snapshot of each; the probability waterfall names the feature and the pillar it came from; every safety and filing flag links to the public record behind it.
- A Verdict tab with the recommendation, the confidence and the leg that capped it
- A probability-of-success waterfall showing each feature’s contribution, labelled directional
- A six-cell operational-risk dashboard, a disproportionality-based safety panel, a sponsor delay ribbon, and a written critique of what the study overstates or has not considered
Read by these expert roles
A role drafts from this pillar; a human reviews and signs off. The sign-off is a 21 CFR Part 11 e-signature on a hash-chained record.
How the roles workHonest limits
What this pillar cannot tell you
Every assessment method has a boundary. Publishing ours is the point — a number you cannot check is worth less than a gap you can see.
Directional, never predictive.
The probability of success is a prior informed by public precedent and adjusted by your trial’s features. It is not calibrated to your programme, it is not a prediction, and it is not a guarantee — and it is labelled directional everywhere it is shown.
The verdict is only as good as its weakest leg.
Combined confidence is the minimum across the analyses it composed, not an average. A thin cohort read caps the whole verdict, and the assessment tells you which leg did the capping instead of hiding it in an aggregate.
Absence of a delay signal is not reassurance.
A private sponsor has no filings to scan, and a public sponsor can be having a bad quarter it has not yet disclosed. An empty ribbon means nothing was found in the public record, not that nothing is wrong.
It is decision support, not the decision.
Nothing here is a medical device, a regulatory submission, or a replacement for your team’s judgement. Every deliverable is drafted for a human expert to review, correct and sign.
The rest of the assessment
No pillar decides alone
Each pillar answers one question. The Verdict composes all six into a single go/no-go, with the risks that could still change it.
See a verdict you can actually check.
Send us a protocol — or just a molecule and an indication. We'll return a fully cited feasibility assessment you can trace, line by line, back to public data — yours to defend in a bid, take to your board or investment committee, or hand to a regulator.