Escaping the Agreement Trap: Defensibility Signals for Evaluating Rule-Governed AI
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866908989090955264 |
|---|---|
| author | O'Herlihy, Michael Català, Rosa |
| author_facet | O'Herlihy, Michael Català, Rosa |
| contents | Content moderation systems are typically evaluated by measuring agreement with human labels. In rule-governed environments this assumption fails: multiple decisions may be logically consistent with the governing policy, and agreement metrics penalize valid decisions while mischaracterizing ambiguity as error -- a failure mode we term the Agreement Trap. We formalize evaluation as policy-grounded correctness and introduce the Defensibility Index (DI) and Ambiguity Index (AI). To estimate reasoning stability without additional audit passes, we introduce the Probabilistic Defensibility Signal (PDS), derived from audit-model token logprobs. We harness LLM reasoning traces as a governance signal rather than a classification output by deploying the audit model not to decide whether content violates policy, but to verify whether a proposed decision is logically derivable from the governing rule hierarchy. We validate the framework on 193,000+ Reddit moderation decisions across multiple communities and evaluation cohorts, finding a 33-46.6 percentage-point gap between agreement-based and policy-grounded metrics, with 79.8-80.6% of the model's false negatives corresponding to policy-grounded decisions rather than true errors. We further show that measured ambiguity is driven by rule specificity: auditing 37,286 identical decisions under three tiers of the same community rules reduces AI by 10.8 pp while DI remains stable. Repeated-sampling analysis attributes PDS variance primarily to governance ambiguity rather than decoding noise. A Governance Gate built on these signals achieves 78.6% automation coverage with 64.9% risk reduction. Together, these results show that evaluation in rule-governed environments should shift from agreement with historical labels to reasoning-grounded validity under explicit rules. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_20972 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Escaping the Agreement Trap: Defensibility Signals for Evaluating Rule-Governed AI O'Herlihy, Michael Català, Rosa Artificial Intelligence Computers and Society Content moderation systems are typically evaluated by measuring agreement with human labels. In rule-governed environments this assumption fails: multiple decisions may be logically consistent with the governing policy, and agreement metrics penalize valid decisions while mischaracterizing ambiguity as error -- a failure mode we term the Agreement Trap. We formalize evaluation as policy-grounded correctness and introduce the Defensibility Index (DI) and Ambiguity Index (AI). To estimate reasoning stability without additional audit passes, we introduce the Probabilistic Defensibility Signal (PDS), derived from audit-model token logprobs. We harness LLM reasoning traces as a governance signal rather than a classification output by deploying the audit model not to decide whether content violates policy, but to verify whether a proposed decision is logically derivable from the governing rule hierarchy. We validate the framework on 193,000+ Reddit moderation decisions across multiple communities and evaluation cohorts, finding a 33-46.6 percentage-point gap between agreement-based and policy-grounded metrics, with 79.8-80.6% of the model's false negatives corresponding to policy-grounded decisions rather than true errors. We further show that measured ambiguity is driven by rule specificity: auditing 37,286 identical decisions under three tiers of the same community rules reduces AI by 10.8 pp while DI remains stable. Repeated-sampling analysis attributes PDS variance primarily to governance ambiguity rather than decoding noise. A Governance Gate built on these signals achieves 78.6% automation coverage with 64.9% risk reduction. Together, these results show that evaluation in rule-governed environments should shift from agreement with historical labels to reasoning-grounded validity under explicit rules. |
| title | Escaping the Agreement Trap: Defensibility Signals for Evaluating Rule-Governed AI |
| topic | Artificial Intelligence Computers and Society |
| url | https://arxiv.org/abs/2604.20972 |