2 Commits

Author	SHA1	Message	Date
Tarik Moussa	3092f1814e	fix(synthesis): namespace lead ids by kind to stop collisions (audit C-11) ... Each stage (find_connections/gaps/improvements/propose_conjectures) numbered its leads from seq=1, so connection L-0001, gap L-0001 and improvement L-0001 all resolved to grounded/L-0001.md. The CLI concatenates them (connections + gaps + improvements) and write_leads writes in order, so later kinds silently overwrote earlier ones — survivors = max(per-kind count), not the sum. Reproduced: 4 distinct leads -> 2 files. _make_lead_id now takes the kind and emits L-C-/L-G-/L-I-/L-X- prefixes, keeping same-seq ids distinct across stages. Regression tests cover the format and the no-collision invariant. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-06-15 11:09:24 +02:00
Tarik Moussa	8cf0cc7e01	feat(synthesis): Lead/Provenance model + grounded leads + conjecture generator ... - codex/synthesis.py: Lead/Provenance dataclasses, find_connections/gaps/improvements, propose_conjectures (status=unverified, quarantined in leads/conjectures/), write_leads (grounded→leads/grounded/, conjectures→leads/conjectures/ HARD invariant) - codex/cli.py: synthesis leads/conjectures/report command group - codex/config.py: F-13 settings (leads_dir, synthesis_llm_*, synthesis_top_k, synthesis_min_grounded_ratio, synthesis_gap_min_coverage) - tests/synthesis/: 42 tests covering model, grounding, conjecture invariant, quarantine, CLI - spike/: F-13 spike script + output (live run attempted) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-06-14 08:37:42 +02:00