analysis-campaign

Perform ablation studies and robustness checks by binding a campaign to the parent run, drafting PLAN.md and CHECKLIST.md files, then launching isolated per-slice runs. This ensures reproducible artifact logging and an aggregated campaign report.

A research campaign plan coordinates disciplined, code-based follow-up analyses after a main result. It uses PLAN.md and CHECKLIST.md to structure ablations, robustness checks, and error analysis into isolated per-slice runs mapped to a paper-facing outline.

Track reproducibility across experiment slices by logging artifacts within isolated per-slice runs. The campaign aggregates these into a final report aligned with your selected outline and evidence contract, updating milestones to guide next steps.

Yes, this applies directly to failure analysis and robustness checks on paper-facing work. It maps isolated per-slice runs to a paper-facing outline and evidence contract, aggregating results into a milestone-updated campaign report.

Aggregate ablation study results by logging artifacts from isolated per-slice runs into a structured campaign. This generates an aggregated campaign report aligned with your selected outline and evidence contract to guide next steps.