Review Console — internal access only
All changes written automatically to drug_assets — from 8-K extractions and OpenFDA polling. Spot-check for errors; use ↩ Undo to revert any bad write.
| Date | Ticker | Drug | Indication | Before | After | Source | |
|---|---|---|---|---|---|---|---|
Summary stats for each completed update.py run — hourly Railway cron and manual runs.
| Execution (ET) | Mode | Companies | Filings | New | → Haiku | → Sonnet | Auto-Applied | Queued | OpenFDA | Errors | Elapsed |
|---|---|---|---|---|---|---|---|---|---|---|---|
Execution — when the run started (ET)
Mode — normal / force-backfill / openfda-only / 8k-only (local if manual)
Companies — total companies scanned
Filings — all 8-K/6-K filings found by EDGAR in the lookback window
New — filings not yet marked was_processed=1 (= Filings − already processed)
→ Haiku — new filings that passed the keyword pre-filter → LLM relevance screen
→ Sonnet — Haiku-screened filings deemed relevant and sent for full extraction
Auto-Applied — updates written directly to drug_assets (high confidence)
Queued — updates routed to proposed_updates for human review
OpenFDA — FDA approvals matched via OpenFDA API
Errors — per-company exceptions during scan
Elapsed — total wall-clock runtime in minutes
Curate the Sunday digest before it sends. Reorder or remove tiles, and add an editorial note that appears at the top of the email. Changes are optional — if you don't touch a week, it sends the computed slate automatically. Target send: .
**bold**, *italic*,
[text](https://…), blank line = paragraph, - bullets.