research-harness-review

You can verify research reproducibility by cross-checking research outputs against logs and study_spec.md without running code. An independent read-only review checks identification credibility, model specification alignment, and numerical accuracy, then generates a review_report.md with a formal APPROVE, REQUEST_CHANGES, or BLOCK verdict.

An independent read-only research review verifies research outputs by reading scripts, logs, and outputs without editing or running code. You need one before releasing research to ensure compliance with study specifications and confirm traceability of all analysis tasks.

You audit research logs by cross-checking outputs with logs and comparing them against analysis_plan.md and study_spec.md. This compliance assurance process flags any deviations from the original study specifications and documents all findings in a structured review report.

No, you do not need to run code to check numerical accuracy and model specification alignment. A read-only review inspects scripts, logs, and outputs directly to verify results, ensuring identification credibility and reproducibility without modifying or executing the analysis code.

The best way to generate a formal review report for research reproducibility is to run an automated review after analysis tasks are complete but before release. It produces review_report.md containing documented findings and a formal verdict of APPROVE, REQUEST_CHANGES, or BLOCK.

A research review returns a BLOCK verdict instead of APPROVE when critical issues are found in identification credibility, model specification alignment, numerical accuracy, or reproducibility. The review_report.md documents all findings and deviations from study_spec.md that triggered the blocking decision.