高效測試時推理的主張層級可靠性評估

摘要

我們提出以「主張層級證偽」作為測試時擴展的原則，並透過「主張層級可靠性評估」（CLR）加以具體實現。CLR 是一個無訓練框架，將測試時運算從額外的解採樣重新分配至針對性驗證。由於整體軌跡評估常因常規詞元造成的訊號稀釋而掩蓋決定性錯誤，CLR 將每個推理軌跡濃縮為一組緊湊且對決策關鍵的主張，藉此隔離其邏輯錨點。此外，考量到在固定模型能力下生成完全正確的解具有內在困難，CLR 將重點轉向語義證偽。此方法利用了「解建構」與「主張反駁」之間的根本不對稱性：建構有效解需要一條無瑕的推理路徑，而反駁錯誤主張僅需找出一個決定性缺陷。這種針對負面證據的搜尋系統性地壓縮了高信心錯誤軌跡的存活空間，並透過非線性可靠性評分有效抑制錯誤共識。在匹配預算下，跨越四個大型語言模型與四個推理基準，CLR 普遍優於 pass@1 與自洽性。例如，在 GPT-OSS-20B/CMIMC25 上，CLR 較 pass@1 高出 27.15 個百分點，並以減少 37.0% 的詞元將自洽性準確率從 77.50% 提升至 82.19%。

English

We propose claim-level falsification as a principle for test-time scaling and instantiate it through Claim-Level Reliability Assessment (CLR), a training-free framework that reallocates test-time compute from additional solution sampling to targeted verification. Since whole-trace evaluation often obscures decisive errors due to signal dilution from routine tokens, CLR condenses each reasoning trace into a compact set of decision-critical claims, thereby isolating its logical anchors. Furthermore, recognizing the inherent difficulty of generating entirely correct solutions under fixed model capabilities, CLR shifts the focus to semantic falsification. This approach exploits a fundamental asymmetry between solution construction and claim refutation. Constructing a valid solution requires a flawless reasoning path, whereas refuting an incorrect claim requires identifying only a single decisive flaw. This targeted search for negative evidence systematically compresses the survival space of high-confidence incorrect traces, effectively suppressing erroneous consensus via nonlinear reliability scoring. Across four LLMs and four reasoning benchmarks under matched budgets, CLR generally improves upon pass@1 and self-consistency. On GPT-OSS-20B/CMIMC25, for instance, CLR exceeds pass@1 by 27.15 percentage-points and raises self-consistency accuracy from 77.50\% to 82.19\% with 37.0\% fewer tokens.