Anti-Gaming 评分完整性

Anti-Gaming Score Integrity

防止 AI 或人类通过作弊手段提升质量分数的完整性保护机制

子问题

1.Target score proximity detection

2.Blind review session isolation

3.Low-score evidence enforcement

4.Wontfix penalty in strict scoring

5.High-score missing justification detection

6.Unverified fix status exploitation

7.Evidence-free score inflation in batch merging

8.Provenance chain tampering for assessment trust escalation

Signals

维度	Desloppify
检测机制	容差带 proximity detection（±5%），warn→penalized 两级响应
评分通道	lenient/strict/verified_strict 三模式并行，wontfix 和未验证修复分级惩罚
证据要求	双向强制：低分(<85)要 finding，高分(>85)要 issues_note，fail-closed 拒绝
盲审隔离	SHA-256 packet hash + runner 白名单 + attestation 短语验证的四层 provenance 信任链
合并策略	Evidence-weighted merge（1+evidence+findings），finding pressure 施加 penalty 和 cap
惩罚机制	penalized 状态重置匹配维度为 0.0，强制 re-review

1.Fail-closed import validation for review findings

2.Evidence-weighted assessment merging

3.Score-independent evidence weighting (1+evidence+findings) prevents hollow high scores from diluting substantive reviews

4.Three-mode scoring channels (lenient/strict/verified_strict) serve different trust levels without score manipulation

5.Bidirectional evidence mandate: low scores need findings, high scores need issue explanations

6.Two-stage gaming response: single match warns, multiple matches reset to zero