arXiv 论文提出 ClaimScope：数学声明审计基准，验证器反馈提升补全率
ClaimScope: Label-Blind Witness Verification for Mathematical Claim Auditing
一篇数学审计方向的论文，搭了个 72 任务的基准，用 Qwen3-8B 测出验证器反馈确实能帮模型交出更多可用反例。做数学推理评估的可以看看它的任务设计。
论文提出 ClaimScope，一个包含 72 个任务的数学声明审计基准，要求模型给出判定、范围、反例与修复。任务覆盖优化器状态传输、有限时域仿射递归和 2×2 线性代数精确计算。在 Qwen3-8B 上的实验显示，验证器反馈在无效声明上接受 23/39 个反例，安全补全 25/72，而无需检查器的通用反馈仅 15/39 和 20/72。判定准确率从 47/72 变为 46/72，表明收益来自可执行证据的结构修正而非更广泛的推理能力提升。
ClaimScope: Label-Blind Witness Verification for Mathematical Claim Auditing
Mathematical agents are commonly evaluated by answer accuracy or proof acceptance. We introduce ClaimScope, a 72-task benchmark for auditing proposed claims through a verdict, scope, counterexample, and repair. Tasks cover optimizer-state transport, finite-horizon affine recurrences, and exact $2\times2$ linear algebra. We evaluate Qwen3-8B with direct generation and four matched two-pass conditions, each starting from the same draft. A witness checker evaluates only the model's submitted state, without consulting reference verdicts or supplying a replacement witness. Its feedback still conveys information about validity: a verified refutation establishes that the claim is false, and rejection reasons can suggest structural fixes. In this evaluation, verifier feedback yields $23/39$ accepted counterexamples on invalid claims and $25/72$ safe completions, compared with $15/39$ and $20/72$ for checker-free generic feedback. Strict completion remains $17/72$; verdict accuracy changes from $47/72$ to $46/72$. The claim-grouped safe-completion difference is $.094$, with a bootstrap 95\% interval of $[.000,.198]$. All seven task-level gains follow horizon/input-length diagnostics. A post-hoc schema-only correction of the direct outputs reaches the same 25 safe tasks without another model call. These results separate claim classification from executable evidence, while locating the observed gains in witness-structure correction rather than demonstrating broader reasoning improvement.
