Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generation

TL;DR

This paper investigates post-hoc rationalization in reverse chain-of-thought generation, formalizes its measurement, analyzes mitigation strategies, and proposes a structural skeleton-guided reasoning method that effectively reduces answer anchoring.

Contribution

It introduces a three-level measurement hierarchy for anchoring, analyzes why semantic suppression fails, and proposes SSR and SSR-D methods to mitigate answer dependence in reasoning traces.

Findings

Semantic suppression increases anchoring at entropic and probabilistic levels.

SSR reduces anchoring across all measurement levels.

SSR-D improves reasoning accuracy by up to 10% over suppression baselines.

Abstract

Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but runs the risk of producing post-hoc rationalizations: when models can see the answer during generation, the answer serves as a cognitive anchor that shapes the entire explanation. We formalize this phenomenon through a three-level measurement hierarchy: lexical, entropic, and probabilistic anchoring, each captures surface artifacts, entropy dynamics, and latent answer dependence, respectively. We analyze semantic suppression, the intuitive mitigation strategy that instructs models to ignore the answer, to find out its counterproduction: while it reduces lexical overlap, it paradoxically increases entropic and probabilistic anchoring. Drawing on Ironic Process Theory from cognitive psychology, we attribute this failure to active monitoring of the forbidden answer, which inadvertently deepens dependence on it. To break this cycle, we propose Structural Skeleton-guided Reasoning (SSR), a two-phase approach that first generates an answer-invariant functional skeleton structure, then uses this skeleton to guide full trace generation. By redirecting the information flow to structural planning rather than answer monitoring, SSR consistently reduces anchoring across all three levels. We further introduce Distilled SSR (SSR-D), which fine-tunes models on teacher-generated SSR traces to ensure reliable structural adherence. Experiments across open-ended reasoning benchmarks demonstrate that SSR-D achieves up to 10% improvement over suppression baselines while preserving out-of-distribution (OOD) generalization.