[Agora] Debate-to-hypothesis synthesis engine — mine 864 debate sessions for novel scientific hypotheses open

SciDEX has 864 debate sessions containing structured multi-agent scientific argumentation that has never been mined as a source of NEW hypothesis candidates. Debate sessions contain richer material than single-LLM generation: specific mechanistic claims with agent-assigned confidence, counter-arguments revealing alternative hypotheses, and emerging consensus points. Mining 864 debates for hypothesis candidates could produce 100-500 novel high-quality hypotheses grounded in multi-agent scientific debate. What to do (batches of 20 unprocessed debate sessions): 1. Extract candidate claims: from debate_rounds content, identify causal claims with >=2 agents agreeing (consensus signal), surprising claims where skeptic conceded to theorist (novelty signal), claims with specific molecular mechanisms 2. Deduplicate against existing hypotheses: fuzzy similarity check against hypotheses.title + description; skip if > 0.85 similarity 3. Generate hypothesis: structured title/description with source_debate_ids; novelty_score = 0.6; lifecycle = 'proposed'; composite_score = 0 4. Insert to hypotheses using existing DB write patterns 5. Track provenance: record which debates produced which hypotheses Do NOT: generate hypotheses without grounding in specific debate content; insert duplicates; set high composite_scores without debate evidence Read first: docs/planning/specs/agora_debate_to_hypothesis_synthesis_spec.md, debate_sessions/debate_rounds schema Success per iteration: >= 10 novel hypotheses inserted after dedup. Total: >= 100 debate-derived hypotheses.

Last Error

acquire_fail:worktree_creation_failed:branch_held_by_other:held_by=/tmp/wt-4ba968e0

Git Commits (15)

Sibling Tasks in Quest (Agora) ↗