{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2611.03627v1",
  "slug": "2611-03627v1-0h0gkar",
  "url": "https://feed7.dev/p/2611-03627v1-0h0gkar",
  "title": "Not All Reasoning Generalizes: A Case Study on Learning Generalizable Reasoning Strategies in RLVR",
  "why_included": "In a controlled RLVR study, models learned structured reasoning in-distribution but switched to ineffective brute force under harder shifts. Co-training with inversion tasks restored some generalization.",
  "summary": "In an equation-solving setup, RLVR models learned step-by-step derivations on familiar tasks but often abandoned them under **out-of-distribution shifts**, such as longer outputs or more solutions. Co-training with bijective inversion tasks helped restore the strategy and accuracy.",
  "practical_implication": "When evaluating reasoning agents, test whether they keep using the intended strategy as task conditions change. In this study, outcome-only rewards could shape reasoning without explicit process supervision.",
  "agent_context": "In an equation-solving setup, RLVR models learned step-by-step derivations on familiar tasks but often abandoned them under **out-of-distribution shifts**, such as longer outputs or more solutions. Co-training with bijective inversion tasks helped restore the strategy and accuracy.\n\nWhen evaluating reasoning agents, test whether they keep using the intended strategy as task conditions change. In this study, outcome-only rewards could shape reasoning without explicit process supervision.\n\nThe results come from a controllable equation-solving framework. The supplied material does not establish whether the intervention transfers to other task types or real-world coding agents.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2611.03627v1",
    "published_at": "2026-11-02T18:53:54.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "benchmark",
  "domains": [
    "research"
  ],
  "topics": [
    "reasoning",
    "agent-evals",
    "agent-reliability"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The results come from a controllable equation-solving framework. The supplied material does not establish whether the intervention transfers to other task types or real-world coding agents."
  ],
  "connected_context": {
    "meaning": "This adds a robustness dimension to reasoning-agent evaluation: success on familiar tasks does not show that a model keeps using its intended strategy when output length or the number of solutions changes. Skill²-Bench complements this by measuring failures when reasoning skills must switch across dependent steps; the study’s equation-solving setup leaves transfer to coding agents unestablished.",
    "corpus_size": 990,
    "generated_at": "2026-11-03T13:56:35.338Z",
    "connections": [
      {
        "title": "Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.05139v1",
        "feed7_url": "https://feed7.dev/p/2608-05139v1-0ji039d",
        "reason": "Skill²-Bench measures skill transitions across dependent steps, while this study tests strategy retention under changed task conditions; together they identify distinct ways reasoning behavior can fail beyond final-answer accuracy."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-11-02T18:53:54.000Z",
  "modified_at": "2026-11-02T18:53:54.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2611-03627v1-0h0gkar",
    "json": "https://feed7.dev/p/2611-03627v1-0h0gkar.json",
    "markdown": "https://feed7.dev/p/2611-03627v1-0h0gkar.md"
  }
}