PharmaCliffBeta status

What's stable, what's beta, what's pending.

No vague "beta" label on the whole product — instead, a component-by-component breakdown. Honest about which pieces are reliable and which are still being tuned with subscriber feedback.

Data ingest — FDA, USPTO, EPO, BigQuerystable

All four registries pull on schedule. Daily Orange Book refresh is verified. INPADOC family expansion across all 14 TA anchors.

Patent index (30,000+)stable

CPC pre-filter against 14 therapeutic-area anchors. New patents flow in continuously; 329 already matched to live Orange Book ANDA routes.

AI rubric v3 — scoringbeta

Five signals, weighted, 0–100 score. Temperature 0 → deterministic. Weights still being tuned against subscriber feedback. v4 expected within a quarter.

Verdict bands (HIGH/MED/LOW)beta

Cutoffs at 75 and 50. Tightening these is the most likely v4 change — current bands let through too much MEDIUM-edge noise.

Weekly digest assembly + sendstable

Tuesday 06:00 UTC. Per-subscriber scoping by tracked areas. Resend delivers; thumbs-up / thumbs-down feedback links per row.

Dashboard — search, watchlistsstable

Sortable search, watchlist save/manage, recent digests view. Sub-second response on Frankfurt Postgres.

Patent detail pagestable

Per-patent view with signals, INPADOC family, Orange Book associations, AI summary. AI summary itself is beta — see below.

AI summary (per-patent)beta

Four summary types: generic opportunity, commercial relevance, competitive landscape, freedom-to-operate. Grounded to patent metadata + abstract only. Hallucination risk minimized but not zero.

API access (v1)pending

REST API with key-based auth exists and is rate-limited. Available on standard tier post-GA. Documentation forthcoming.

What changes vs. what's locked

The contract around beta.

What can change without notice: the AI rubric (signal weights, verdict cutoffs), the sample digest layout, the dashboard component arrangement, anything cosmetic.

What we won't change without 90 days' notice: the price for any founding subscriber (€39/mo, locked for life), the data sources, the weekly delivery cadence, the public methodology page.

What we won't do, period: silently re-score past digests (they stay at the rubric version they were sent under), share subscriber data with third parties, add upsells without a price-change notice, switch from public registries to a paid data source without disclosing why.