185 target dossiers exist with confidence tiers (91 tier_1_high, 75 tier_2_medium, 19 tier_3_exploratory) and 2,215 hypothesis-target links connecting 1,305 hypotheses. BUT the `recommendation`, `mechanism_of_action`, and `existing_compounds` columns are NULL for ALL 185 targets.
**Why this matters**: Task `14954567` (hypothesis-target relevance scoring) just shipped 2,215 hypothesis-target links. The evidence is assembled. The missing step is synthesizing it into actionable drug discovery output: "pursue target X with mechanism Y, supported by these validated hypotheses, these compounds create/avoid this liability."
**What success looks like**:
- ≥50 tier-1-high targets have `recommendation` (≥200 words: strategy, mechanism, evidence basis, existing compounds, confidence assessment, key risks)
- ≥50 have `mechanism_of_action` (1-3 sentence mechanism summary)
- ≥50 have `existing_compounds` (jsonb array: compound names, status, indication)
- Each recommendation cites top 3 linked validated hypotheses by composite_score
**What to read first**:
- `docs/planning/specs/quest_atlas_target_recommendations_spec.md` — full spec (in branch 80ffb77b-cycle6-specs)
- Query `target_dossier JOIN hypothesis_target_links JOIN hypotheses WHERE status='validated'` for evidence
- Check `targets` table for gene_symbol, disease, druggability metadata
**What NOT to do**: Do not hallucinate compound names. Do not modify api.py. Focus tier-1 first.
Last Error
watchdog: worker lease expired; requeued
Git Commits (1)
[Atlas] Generate drug target therapeutic recommendations for tier-1 neurodegeneration targets [task:1544b567-7581-467f-9a9e-1605913998b9] (#1340)2026-04-28