Audit experimental artifacts for provenance, normalization, and result integrity.
npx skills add https://github.com/Mang30/myskills --skill experiment-audit-mang30
Please help me install this Agent Skill. Skill: experiment-audit Source: https://github.com/Mang30/myskills/tree/main/skills/experiment-audit Command: npx skills add https://github.com/Mang30/myskills --skill experiment-audit-mang30
This Skill reduces the risk of publishing invalid or misleading experimental results by detecting common integrity failures like fake ground truth, self-referential score normalization, phantom/mismatched result claims, and scope overreach.
Use the experiment-audit skill with your experiment directory or results path to generate EXPERIMENT_AUDIT.md and EXPERIMENT_AUDIT.json for an integrity verdict.
High-intent search queries and answers about installing and using this skill.
You can verify experiment results by auditing experimental artifacts to validate ground-truth provenance, score normalization, result-file existence, and scope alignment. This generates a structured PASS/WARN/FAIL report and JSON verdict.
Detecting fraudulent research reporting involves checking for fake ground truth, self-referential score normalization, phantom result claims, and dead-code coverage. A cross-model review reads scripts, metrics, and trackers in a read-only sandbox to identify these integrity failures.
Cross-model evaluation audit separates roles: an executor collects file paths without summarizing, while a separate reviewer model reads files in a read-only sandbox to judge integrity patterns and produce the final verdict.
To check evaluation scope alignment, audit your experimental scripts, metrics, trackers, and paper narrative to ensure the evaluation scope matches how the paper describes it. This prevents scope overreach in research claims.
Yes, you can automate experiment integrity checks by running an audit that validates ground-truth provenance, score normalization, result-file existence, and dead-code coverage across your experiment directory. It outputs both a human-readable report and a machine-readable JSON summary for downstream pipeline integration.
Shallow-WAudit completed experiment workflows for integrity issues and produce PASS/WARN/FAIL evidence.
lightrain-aValidates experiment integrity by checking provenance, metrics behavior, file existence, and produces audit reports with verdicts.
jkfeeAudit experiment integrity across evaluation scripts, results, and paper claims.
goupup-aiAudit machine learning experiment results for fraudulent reporting patterns.
