{
  "id": 1437,
  "url": "https://arxiv.org/abs/2606.05932v2",
  "title": "A Pre-Registered Causal Partition of Self-Consistency Elicitation and Reward Design in RLVR",
  "summary": "Reinforcement learning from verifiable rewards (RLVR) improves reasoning even when the reward signal is spurious -- assigning credit to the group-plurality answer rather than a ground-truth verifier. Practitioners commonly interpret naive = acc(TRUE) - acc(RANDOM) as the reward-design effect. We prove this estimand is systematically biased: it conflates self-consistency elicitation (sharpening the policy toward its modal answer via majority pseudo-reward) with genuine reward-design signal. Using",
  "authors": "Yuze Gao",
  "category": "research",
  "topics": "bias-fairness,regulation",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-04T09:35:54.000Z",
  "fetched_at": "2026-07-14T14:15:17.101Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/1437",
  "original_url": "https://arxiv.org/abs/2606.05932v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}