experiment-section-auditor

Experiment Section Auditor

Scope

Use this skill when the user is designing, revising, or writing the experiment section and wants a constraint layer — not an author. It covers two things and only these two:

• experiment and ablation design: whether the study actually argues the paper's main line, and what the minimal set is under real resource limits;
• results narration: whether results are written as argument tied to claims rather than as a report of numbers.

Read references/experiment_section_guide.md for the claim–evidence mapping method, the minimal-set-under-constraints procedure, the ablation discipline, the report-style-versus-argument checklist with rewrite examples, and the output template. For a quick check, the rules below are enough.

Use another skill when the task is primarily:

• how the contribution — including the one-line results summary in the abstract or introduction — is framed: use paper-argument-reconstructor (if available). Pull the paper's claims from there and audit the experiments against them. That skill owns the abstract/intro wording; this one owns the results narration inside the experiment section. They must stay consistent: the abstract's results sentence must not claim more, or under looser conditions, than the section supports;
• whether the idea is novel or what can be claimed at all: use idea-novelty-auditor (if available). That fixes the defensible claims; this checks the experiments that support them;
• TeX-safe or final-language polishing of the section: use paper-polisher (if available);
• choosing a target journal: use journal-recommender (if available).

This skill does not lead the experiment design and does not write the section for the user. It audits, flags, and proposes the minimum needed to support the claims. The user drives.

Verification Mode

This skill is audit-first in both modes; what changes is how much it proposes.

• Default (the user's own plan or draft): audit against the rules below, flag gaps and report-style narration, and propose only the minimal experiment or ablation needed to close each gap. Do not rewrite the section or design a full campaign.
• Strict verification (an experiment plan or narration another tool or skill already produced): do not propose a competing design; check claim-coverage, feasibility assumptions, over-claims, and report-style, and report only the deviations. Fix one in place only if it breaks a non-negotiable rule (a fabricated result, an experiment assumed feasible or presented as done, an invented anomaly cause, or an unconditional over-claim).