PharmaCliffHow it works

From four public registries to one Tuesday digest.

A five-stage pipeline. Every stage is disclosed, the rubric weights are published, the source list is on the methodology page. No black-box scoring, no enterprise-only API.

Step 01

Ingest

Pull from FDA Orange Book daily, USPTO PAIR weekly + on-demand, EPO INPADOC weekly families, Google Patents BigQuery for historical backfill.

4 public registries
Step 02

Expand

INPADOC family expansion adds IN, CN, BR coverage when a family member is present. Cross-jurisdiction cliffs surface as a single signal.

~30,142 patents indexed
Step 03

Score

Five signals — Orange Book route, family convergence, claim breadth, time-to-expiry, TA-anchor depth — weighted into a 0–100 score at temperature 0.

Same input → same score
Step 04

Bands

HIGH ≥ 75 · MED 50–74 · LOW < 50. Verdict bands are published; subscriber feedback tunes the next rubric version.

Beta AI · rubric versioned
Step 05

Send

Tuesday 06:00 UTC. 24 ranked patents per subscriber, scoped to their tracked therapeutic areas. Full archive in dashboard.

Weekly · 06:00 UTC
Cadence

The weekly rhythm.

Monday 22:00 UTC: indexes finalize, scoring runs, the digest is assembled for every active subscriber scoped to their tracked therapeutic areas.

Tuesday 06:00 UTC: emails go out. Mid-tier generics teams in EU + India see them at 08:00 EET / 11:30 IST — first thing in the working day.

Throughout the week: subscriber thumbs-down flags on wrong verdicts are reviewed and used to tune the next rubric version. Dashboard is open for search, watchlists, and patent-detail drill-downs at any time.

End of month: rubric changes (if any) are published in the methodology changelog with rationale. Past digests are NOT silently re-scored — they stay at the rubric version they were sent under.