[Agora] Auto-trigger debates for low-quality or conflicting artifacts done analysis:5

Goal

Automatically identify artifacts that would benefit from debate and trigger artifact-review
debates. Targets include: artifacts with low quality scores, conflicting evidence,
high usage but no debate history, and newly-extracted experiments with methodology concerns.

Acceptance Criteria

- Quality score < 0.4 AND usage count > 3 (widely used but low quality)
- Evidence balance < -0.3 (more contradicting than supporting evidence)
- High usage (top 10%) AND zero debates (never scrutinized)
- Conflicting replication status (experiments)
- Quality score dropped > 0.2 in last 30 days

Dependencies

• agr-ad-05-PROF — Quality profile signals inform trigger decisions

Dependents

• None (leaf task)

Work Log

2026-04-25 — Implementation

What was done:

scidex/agora/debate_trigger.py (new file, 455 lines):

- find_debate_candidates(limit=20) — ranked list of DebateTrigger objects
- 5 trigger rules: low_quality_high_usage, evidence_imbalance, high_usage_no_debate, conflicting_replication, quality_drop
- Deduplicates by artifact_id, sorts by trigger_score descending
- Rate-limited to MAX_AUTO_DEBATES_PER_DAY=3 via knowledge_gaps count with [AUTO_TRIGGER] tag
- trigger_artifact_debate() — creates gap with TARGET_ARTIFACT + AUTO_TRIGGER tags
- _get_debate_count() — counts debates via artifact_debates + debate_sessions

scidex/agora/scidex_orchestrator.py (orchestrator integration):

- Parses [TARGET_ARTIFACT type=X id=Y] from gap description at start of run_debate
- Sets target_artifact_type/target_artifact_id on debate_sessions INSERT
- After debate completes: _update_artifact_debate_outcome() — records in artifact_debates, updates support/contradiction counts, creates artifacts_history entry for quality drops

api.py — added GET /api/debate/candidates endpoint returning ranked DebateTrigger list

Status: Committed and pushed to orchestra/task/agr-ad-0-auto-trigger-debates-for-low-quality-or (commit 1855f6871)

Note on data: Rule 2 (evidence_balance < -0.3) returns 0 candidates (no artifacts have 3+ debate entries with negative balance yet). Rule 4 (conflicting replication) returns 0 (no conflicting experiments currently). Rule 1 returns 0 (no artifacts have quality_score < 0.4 AND usage signals). Active candidates: high_usage_no_debate (top-10% usage, 0 debates) for 20 wiki pages.

Verification — 2026-04-25 19:50 UTC

What was verified:

Code compiles: python3 -m py_compile scidex/agora/debate_trigger.py and scidex_orchestrator.py — both OK

Function works: find_debate_candidates(limit=5) returns 5 candidates (high_usage_no_debate rule, wiki pages with top-10% usage and 0 debates)

API route exists in api.py at line 18294: @app.get("/api/debate/candidates") with correct docstring

Orchestrator integration: parses [TARGET_ARTIFACT type=X id=Y] and _update_artifact_debate_outcome() after debate completes

Rebase: confirmed branch is up to date with origin/main (diverged by 2 commits that are the implementation itself)

All acceptance criteria addressed by the implementation

Commit: 1855f6871 (auto-trigger debates for low-quality or conflicting artifacts)