The platform
Public data in. A signed, replayable verdict out.
Six layers, in order. Each one either produces something checkable or refuses to produce anything — because the value of this system is not that it answers quickly, it is that the answer survives being checked.
Every claim bound to a source record and a snapshot
The pipeline
Six layers, in order
01 · Ingest
Where does the evidence come from?
Around 50 live public-data connectors are run for your study — not one static database licensed years ago and left to age.
A connector is a maintained link to a single public source. It knows that source’s query language, its field names, its rate limits and its publication cadence, and it is kept working as the source changes. When a study is created, the engine runs the connectors that matter for that indication, intervention and phase, and records which ones returned data and which came back empty. Seven registries and databases form the backbone; national trial registries, patent, pricing and filing sources sit around them.
A live query, not a copy
Each connector reads the source itself and pins what it read to a snapshot, so the record on your page is the record that existed at that moment — not an extract of unknown age.
Seven backbone sources
ClinicalTrials.gov, AACT, PubMed, FDA, EMA, Open Targets and ChEMBL carry most of the assessment. The rest of the connectors surround them — national registries worldwide, patents, drug pricing, and sponsor filings.
An honest evidence ledger
The run records which sources returned data for your study and which did not. A gap is reported as a gap rather than quietly filled in.
02 · Ground
How does a number become checkable?
Extraction turns a source record into a structured fact; grounding binds every claim to the exact record and snapshot it came from.
Nothing in the assessment is allowed to float. When the engine reads a value out of a source — an enrollment figure, a label section, a binding constant, a filing sentence — it stores the value together with the source it came from, the query that found it, and the snapshot it was read from. The claim that later appears in your assessment carries that binding. That is the whole reason a number on the page can be opened, traced and re-checked long after the meeting that used it.
Value, query, snapshot
Every extracted fact keeps the three things you need to reproduce it: what was read, how it was found, and when it was read.
Claims are bound, not summarised
A sentence in a deliverable points at the record behind it. A claim the engine cannot bind to a record does not ship as a fact.
Graded, not polished
Each analysis carries an evidence grade, from single-source up to triangulated, so a number built on thin evidence says so on its face instead of borrowing the confidence of the ones that are not.
03 · Analyse
What actually gets computed?
More than 30 analyses across six pillars — the same structured work-up for every study, computed from grounded evidence.
The pillars are fixed, so two studies can be compared and one study can be re-run. Each pillar answers one question a decision actually turns on, and each analysis inside it is computed from the grounded evidence rather than asked of a model in free text. Where an analysis cannot be supported by the evidence that came back, it says so instead of producing a confident number.
Rationale — should this drug work?
Target validation · Binding affinity · Off-target & interaction risk · Structural viability
Design — is the trial designed to win?
Competitive landscape · Why comparable trials failed · Design vs. indication norm · Statistical power · Regulatory-precedent timeline · Protocol-amendment risk
Cohort — can you find the patients?
Cohort sizing · Eligibility complexity · Real-world cohort
Where & How — where do you run it, and at what cost?
Study sites & contacts map · Site & investigator selection · Site density vs. competition · Geographic prioritisation · Cost feasibility · Operational benchmarks
Commercial — is it worth it if it works?
Market sizing (TAM / SAM / SOM) · Risk-adjusted valuation (eNPV) · Value-based price · Standard of care · Patent landscape & exclusivity
Verdict — go, or no-go?
Go / No-Go recommendation · Probability of success · Operational-risk dashboard · Adverse-event signal · Sponsor delay signals (SEC filings) · Verdict interpretation
04 · Draft
Who turns the analysis into a document?
Seven expert role workspaces write the documents your decision needs — drafted from the analyses, with citations, never presented as final.
The assessment answers the question; the deliverable is what circulates. Each role workspace reads the analyses relevant to it and drafts from them — the biostatistics workspace from the power and cohort work, the commercial workspace from market sizing, price and patent runway. A draft that cannot point at the analysis behind a claim does not make the claim.
Protocol Design
Drafts the trial protocol synopsis from the verdict, design and cohort analyses — every claim cited.
Commercial Strategy
Market size, standard of care, value-based price, patent runway, and risk-adjusted valuation scenarios.
CMO / VP R&D
Portfolio-level valuation, decision-gate memos, and readout-risk framing for the go/no-go call.
Biostatistician
Statistical analysis plans, sample-size estimates, and data-monitoring-board materials.
Clinical Pharmacologist
Dose-selection memos, exposure–response modelling, and drug-interaction risk.
Study Start-up
Country activation tracking and contract groundwork for the sites the assessment ranked.
CRO Commercial
Bid-pipeline and portfolio views for CRO business development teams.
05 · Govern
Who is accountable for the answer?
Your team reviews and signs. Sign-off is a 21 CFR Part 11 electronic signature, and the run lands on a hash-chained audit trail.
The platform drafts; a named person approves. Nothing is treated as decided because the engine produced it. Each approval is captured as an electronic record and signature and pinned to the exact evidence it was based on, and every run writes an append-only audit record where each entry references the cryptographic hash of the entry before it — so a retroactive edit breaks the chain and is detectable rather than invisible.
Nothing is approved by the machine
Every deliverable is reviewed by your team before it counts. The platform can draft a recommendation; it cannot sign one.
21 CFR Part 11 e-signature
Sign-off is captured as an electronic record and signature — who approved it, when, and against which version of the evidence.
Hash-chained audit trail
Inputs, citations, scorer versions and outputs land on an append-only chain. Each entry carries the hash of the one before it, so tampering is detectable after the fact.
Re-review when the evidence moves
If the underlying evidence later changes, the signed decision is automatically flagged for re-review. Nothing stays approved on stale data.
Your data never trains a model
Your protocols, documents and studies are processed to answer your question. They are not used as training data.
06 · Deliver
What do you actually get?
A six-pillar assessment and a go/no-go verdict where every number opens onto the public record it was built from.
The output is not a score with a paragraph attached. It is a recommendation, the reasoning that produced it, the risks that could still change it, and a citation on every figure — readable by the clinical, commercial and financial people in the same room, and defensible to a board, a sponsor or a regulator months later.
The verdict, with its reasons
A go/no-go recommendation alongside the operational risks, adverse-event signals and sponsor delay signals that could move it.
Every number opens
Click a figure and you get the source, the query and the snapshot behind it, so your team can confirm the claim independently rather than trusting the engine.
Directional — and it says so
Probability of success frames and ranks risk against public precedent. It is not a calibrated forecast, not a prediction, and not a guarantee, and the assessment states that where it appears.
Chain of custody
What is recorded, and what you can check.
A claim you cannot re-open is a claim you have to take on trust. At every layer the system records something specific, and every one of those records is inspectable later.
| Layer | What is recorded | How you check it |
|---|---|---|
| Ingest | The connector queries the public source and records what came back — and what did not. | Which sources were pulled for your study, and which returned nothing. |
| Ground | The figure is stored with the record it came from, the query that found it, and the snapshot date. | The exact public record the number was read from. |
| Analyse | The cohort analyses compute the figure and attach an evidence grade to it. | The grade, and the grounded inputs the calculation used. |
| Draft | A role workspace writes the figure into the memo, carrying its citation with it. | Which analysis each sentence in the document came from. |
| Govern | A named reviewer signs the memo; the run lands on the hash-chained audit trail. | Who approved it, when, and against which version of the evidence. |
| Deliver | The figure appears in the assessment as a citation you can open. | The whole chain — months after the meeting that used it. |
Honest limits
What the platform does not do
Stated plainly, and before you ask. A vendor who will not name the boundary is asking you to find it yourself, in production.
It does not touch patient-level data
The engine runs on public records and protocol-level inputs. It does not read your EHR, your EDC, or any identifiable patient record, and it does not need to in order to produce an assessment.
It does not replace a qualified professional
Every deliverable is a draft for your clinical, statistical, regulatory and commercial experts to review, correct and sign. The platform is decision support, not a decision-maker, and not medical or investment advice.
Probability of success is directional
It frames and ranks risk against public precedent so you can compare options and see what drives the number. It is not a calibrated forecast, not a prediction, and not a guarantee of any trial outcome.
It cannot see what was never made public
Unpublished results, confidential competitor programmes and unregistered trials sit outside every public source. Where the evidence is thin or missing, the assessment reports the gap rather than filling 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.