Start here
How to read an assessment
The six pillars, the questions each one answers, and how to tell a number backed by several independent sources from one resting on a single record.
An assessment is more than thirty analyses organised into six pillars. Each pillar answers one question a real go/no-go conversation turns on, and each analysis inside it shows its working.
The habit worth building is simple: read the number, then read the band next to it. A figure with thin evidence behind it says so on its face, and is meant to be treated differently from one that several independent sources agree on.
The six pillars#
The pillars run in the order a decision actually gets made — from whether the science holds up to whether the whole thing is worth doing.
| Pillar | The question | What you will find |
|---|---|---|
| Rationale | Should this drug work? | Target validation, binding affinity, off-target and interaction risk, structural viability. |
| Design | Is the trial designed to win? | Competitive landscape, why comparable trials failed, design versus the 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 and contacts map, site and investigator selection, site density versus 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 and exclusivity. |
| Verdict | Go, or no-go? | Go/No-Go recommendation, probability of success, operational-risk dashboard, adverse-event signal, sponsor delay signals from filings, and an interpretation of the call. |
Alongside the six pillars, the study carries an Evidence surface listing every source the run touched, a Snapshots surface holding the exact records the run read, and a Tasks surface where the deliverables and sign-offs live.
Evidence bands — how much to trust a number#
Every analysis carries an evidence band derived from how many independent sources support it and how much they agree. It is the fastest honest read on any figure in the assessment.
- No data
- Nothing backed this number. Do not act on it.
- Weak
- Very little evidence. Directional at best.
- Single source
- One connector supports this. Not yet corroborated.
- Converging
- Multiple sources point the same way.
- Triangulated
- Several independent sources agree. The strongest evidence tier in the system.
A band is a statement about the evidence, not about the outcome. A triangulated number can still be wrong about your trial — it just is not wrong because we made it up.
Basis — where each input came from#
Modelled figures also carry a basis, which says where the input behind them came from. This is what separates a fact you supplied from a constant we chose.
- Stated — you entered it.
- Published — a cited figure from the literature or a regulatory record.
- Computed — derived from trial-registry records for this indication and phase.
- Curated — a hand-picked constant that carries a citation.
- Heuristic — a constant that does not carry one.
- Assumption — a modelling knob, not a measurement.
- Not modelled or absent — no factor was applied, or the input does not exist on this study.
If a headline number rests on a heuristic or an assumption, you can see that before you quote it in a board paper.
Reading the probability of success#
Probability of success is a base rate: the historical share of comparable trials that cleared this phase. It is directional evidence about a class of programmes, not a forecast for yours.
- It is not calibrated against your programme, and we do not present it as a measured probability of your trial succeeding.
- The assumptions and comparator set behind it are on display, so you can disagree with them specifically rather than in general.
- Use it to rank options and to argue about the drivers — not as a number to put in a term sheet.
Confidence scores in the product work the same way. While uncalibrated, a confidence score orders results — more evidence means higher — but it is not a percentage chance of being right.
Checking a number you do not believe#
Disbelief is the intended reaction to an unfamiliar figure, and the whole product is built so you can act on it in about thirty seconds.
Open the citation
Every figure carries the public source it came from, the query that produced it, and the date the record was read.
Go to the record
The citation links back to the public record itself. Read the trial registration, the label, the paper.
Check the band and the basis
If the number is single-source or rests on an assumption, you have found the seam — and the assessment already told you where it was.
Correct the input
If the study definition was wrong, fix it and re-run. Analyses are deterministic and re-runnable, so a corrected input produces a corrected assessment rather than a differently-worded one.
Related
Next
The evidence
Citations & the audit trail
What a citation contains, how the hash-chained Part 11 record makes a run tamper-evident, and how to re-open a verdict long after the meeting that used it.
OpenThe evidence
Limits & caveats
The honest boundaries: what public data cannot tell you, which numbers are upper bounds, and where a human has to make the call.
OpenStart here
Start a study
What the platform needs from you, what it works out for itself, and what happens between the moment you press Confirm and the moment a verdict appears.
OpenSee 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.