{
  "schema_version": "1.0",
  "id": "archive:https://arxiv.org/abs/2607.11849v1",
  "slug": "2607-11849v1-1sjdl4n",
  "url": "https://feed7.dev/p/2607-11849v1-1sjdl4n",
  "title": "AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification",
  "why_included": "AdvancedMathBench separates proof writing from verification and finds frontier models especially weak at rejecting invalid proofs, a warning against trusting agent self-review on rigorous reasoning.",
  "summary": "AdvancedMathBench separates construction from checking: **ProverBench has 296 problems** from undergraduate through doctoral qualifying-exam level, while **VerifierBench has 888 generated proof trajectories** with expert ground truth.",
  "practical_implication": "For reasoning-heavy agents, evaluate generation and verification independently. The best reported proof scores were **75.8 on UGD and 66.1 on QE**, so adding a verifier does not by itself establish correctness.",
  "agent_context": "AdvancedMathBench separates construction from checking: **ProverBench has 296 problems** from undergraduate through doctoral qualifying-exam level, while **VerifierBench has 888 generated proof trajectories** with expert ground truth.\n\nFor reasoning-heavy agents, evaluate generation and verification independently. The best reported proof scores were **75.8 on UGD and 66.1 on QE**, so adding a verifier does not by itself establish correctness.\n\nThe best verifier reached only **65.1 Balanced F1**, and models generally had low true-negative rates. The benchmark targets advanced mathematics, so its results do not directly measure coding-agent review.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2607.11849v1",
    "published_at": null
  },
  "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 best verifier reached only **65.1 Balanced F1**, and models generally had low true-negative rates. The benchmark targets advanced mathematics, so its results do not directly measure coding-agent review."
  ],
  "lifecycle": "Current",
  "published_at": null,
  "modified_at": null,
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2607-11849v1-1sjdl4n",
    "json": "https://feed7.dev/p/2607-11849v1-1sjdl4n.json",
    "markdown": "https://feed7.dev/p/2607-11849v1-1sjdl4n.md"
  }
}