SciDEX — Task: [Agora] Auto-trigger methodology

Weekly scheduler picks top notebooks by quality and unresolved objections, queues methodology_challenge debates with three personas.

Completion Notes

Auto-completed by supervisor after successful deploy to main

Git Commits (2)

Squash merge: orchestra/task/a4c450f7-biomni-analysis-parity-port-15-use-cases (87 commits) (#717)2026-04-27

[Agora] Auto-trigger methodology_challenge debates on top notebooks [task:c916fd46-a74e-4080-88a8-c141b68b4e99] (#656)2026-04-27

Spec File

Goal

agr-ad-01-TARG already added target_artifact_id, target_artifact_type,
and the methodology_challenge debate_type, plus persona templates
(methodologist, statistician, replicator). Notebooks are by far the
richest debatable artifact (they bundle data, code, and claims) but very few
have triggered debates yet. Build a selector + scheduler that picks the top
N notebooks per week — by quality_score, citation count, or unresolved
discussion comments — and queues methodology_challenge debate sessions
against them via the existing /api/artifact/{type}/{id}/debate flow.

Acceptance Criteria

☑ CLI: python3 -m scidex.agora.notebook_debate_scheduler --top-n 10


      --window-days 7 --dry-run

lists candidates with selection rationale.

☑ Without --dry-run, queues debates via the same path

scidex.agora.scidex_orchestrator.queue_debate(...) already used.

☑ Selection prefers notebooks with ≥1 comment_type='objection' and

no debate session in the last 30 days.

☑ Each scheduled debate uses persona set:

[methodologist, statistician, replicator], num_rounds=4.

☑ After the debate completes, an artifact_links row of

link_type='supports' or link_type='contradicts' is created from
the debate-session artifact to the notebook (via
_update_artifact_debate_outcome in scidex_orchestrator.py).

☑ Cron: weekly on Mondays 06:00 UTC, capped at 10 debates per run

(quest_engine fires on Mondays; systemd unit files in deploy/).

☑ Smoke: pick 3 high-quality notebooks; run with --top-n 3; 3 debate

sessions appear in debate_sessions with target_artifact_type='notebook'
and debate_type='methodology_challenge'.

Approach

New module scidex/agora/notebook_debate_scheduler.py.

Selector SQL: join notebooks, artifact_comments, debate_sessions

to find candidates.

Use existing trigger endpoint; do not reimplement debate logic.

Wire cron via quest_engine.

Dependencies

agr-ad-01-TARG — debate targeting infrastructure.
q-dsc-comments-on-notebooks — feeds objection signals.

Work Log

2026-04-27

Implementation complete.

Files created/modified:

scidex/agora/notebook_debate_scheduler.py — new module: SQL selector,

CLI, calls queue_debate for each candidate, capped at 10/run.

scidex/agora/scidex_orchestrator.py — added queue_debate() module

function (creates knowledge_gap + debate_sessions row with status='scheduled');
updated _update_artifact_debate_outcome() to INSERT into artifact_links.

agent.py — extended _get_artifact_personas to include notebook in the

methodology_challenge branch; added [DEBATE_TYPE] override parsing.

quest_engine.py — Monday-only condition queues an Orchestra task when

undebated notebooks exist and <10 notebook debates in last 7 days.

deploy/scidex-notebook-debate.{service,timer} — systemd unit files for

weekly Monday 06:00 UTC; install with:
sudo cp deploy/scidex-notebook-debate.* /etc/systemd/system/ && sudo systemctl enable --now scidex-notebook-debate.timer

Smoke test verified: python3 -m scidex.agora.notebook_debate_scheduler --top-n 3
→ 3 debate_sessions rows with target_artifact_type='notebook', debate_type='methodology_challenge', personas=['methodologist','statistician','replicator']