What problem does it solve? After a main experiment produces a result, research quests often need structured follow-up runs such as ablations, sensitivity checks, or error analysis. Without a disciplined protocol, these runs become an untracked pile of extra experiments with buried negative results and no clear conclusion. ## Core Features & Use Cases - Campaign Charter & Planning: Creates durable PLAN.md and CHECKLIST.md files that bind every analysis slice to a research question, hypothesis, and falsifiable expectation before any run launches. - Isolated Slice Execution: Splits the campaign into one-question-per-run slices with controlled variables, smoke tests for long runs, and managed bash_exec monitoring with stall detection. - Paper-Facing Integration: Maps each slice to a paper experiment matrix row with section_id, claim_links, and paper_role so completed evidence flows back into outlines and evidence ledgers. - Use Case: After a main training run shows a gain, use this Skill to run a claim-critical ablation first, then robustness checks across seeds, record each slice as a run artifact, and aggregate a report classifying the outcome as stable support, partial support, contradiction, or unresolved ambiguity. ## Quick Start Launch an analysis campaign for the accepted main run with one ablation slice and one robustness slice, then aggregate the results into a campaign report.
