Pipeline Atlas · a drug-landscape explorer
The map that refuses to guess
A browsable index of clinical drug development — decades of trials grouped by modality, target, mechanism, sponsor, and phase, with the evidencing trials one click away — built on one constraint that outranks every feature: it must never present an absence as a finding.
live a sketch of the idea, computed on this page — programs drifting through phases, bright where provenance is resolved, faint where inferred; hatched cells are views the app withholds rather than draws from partial data. Not the app's data.
The problem it works on
The public trial registry is twenty-five years of sponsor-authored free text, in which the same molecule appears under five different names across its lifecycle and the most interesting attributes are simply absent for most rows. Everyone who builds on it faces the same temptation: fill the gaps plausibly, so the map looks complete. Atlas is a bet on the opposite — that a map with honest holes supports better decisions than a map that fills them.
How it stays honest
Every displayed attribute carries a provenance tier — resolved, annotated, inferred, or unknown — and a value is empty if and only if its tier is unknown. A statistic mechanically inherits the weakest tier among its inputs; the pipeline fails loudly rather than clamping silently past a gap. Views whose underlying coverage falls below a measured threshold are withheld outright, with the coverage figure stated — in the August 2026 snapshot, built from 220,376 studies and 766,867 intervention rows, 97 collections render fully, 81 with a caveat, and 20 are withheld. The most instructive defect in its history is preserved as a lesson in the docs: a fuzzy-matching rule once read a two-letter registry abbreviation as a drug name and wrote 433 confident fictions at the highest tier — all purged to become honest absences, because confident errors cost more than the gaps they fill.
See it working
The app is in private beta while coverage grows. If the landscape view, the provenance machinery, or the honest-holes discipline is relevant to your work, ask for a walkthrough — the form below opens a pre-addressed email in your own mail app; nothing is sent from this page and nothing about you is collected by it.
Independent project; public data only. Nothing in it is medical, scientific, or investment advice.
What this does not claim
- That an absence is a finding. Where the registry cannot support a view, the map draws it as withheld rather than guessing.
- That it is advice. It is a data explorer over a public registry — not clinical, medical or investment guidance — and it is in private beta.
Built and checked under the standard.