id: experimental-design-standards
name: Experimental claims meet design standards
description: >
  Any section presenting experimental results must clearly state: (1) whether
  conditions were pre-registered or post-hoc; (2) N per condition; (3) whether
  the auditor/evaluator is independent of the treatment generator (same-family
  AI auditing same-family AI output is a confound that must be disclosed);
  (4) whether statistical inference is claimed and if so whether sample size
  supports it. Post-hoc conditions designed to fix gaps found in prior runs
  must be labeled as iterative product development, not experimental evidence.
  Pass: all conditions clearly labeled, confounds disclosed, inference claims
  calibrated to N. Fail: post-hoc conditions presented as experimental evidence
  without explicit label.
property: Verifiable
tags: [academic-paper]
phase: staging
trigger: pr
blocking: true
check:
  type: manual