Reproducing and Stress-Testing Two Approaches to LLM Reasoning Reliability: Test-Time Probability Aggregation and Logic-Representation Editing

• URL: http://arxiv.org/abs/2608.08514v1
• Date: Sun, 09 Aug 2026 06:22:02 GMT
• Title: Reproducing and Stress-Testing Two Approaches to LLM Reasoning Reliability: Test-Time Probability Aggregation and Logic-Representation Editing
• Abstract summary: We independently reproduce two methods for making large language model (LLM) reasoning more reliable, and stress-test them across domains and models.<n>The first, RPC, aggregates token probabilities and self-consistency at inference; the second, LCF, trains projectors that split hidden states into "content" and "logic" and edits the logic part toward a valid region.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We independently reproduce two recent methods for making large language model (LLM) reasoning more reliable, and stress-test them across domains and models (RPC across four new task domains with Qwen3-8B, LCF across four 7-8B models). The first, RPC, aggregates token probabilities and self-consistency at inference; the second, LCF, trains projectors that split hidden states into "content" and "logic" and edits the logic part toward a valid region. Validating such reliability claims matters because the original evaluations are run by each method's own authors and were never independently reproduced or stress-tested across models and domains, and LCF shipped no public code. We re-run RPC's published-path aggregation and re-implement LCF's projector, contrastive, and intervention pipeline, then extend both to text-to-SQL, legal extraction, fallacy identification, and precedent grading, and probe LCF's representation directly. RPC reproduces the original grid exactly on the authors' released reasoning paths; on four new domains its edge over self-consistency is never significant (ties or small mixed differences, paired p >= 0.28), and on BIRD, the one domain where we vary the budget, the edge grows with K as predicted but its largest gap (+2.5 accuracy at K=32, p=0.16) reverses to -0.25 when we enlarge the sample to n=200. LCF's logic-validity direction is real but weak (0.82 separability at the single best sub-layer versus 0.95 for a semantic-attribute control); its one positive effect (Qwen3 $Δ$Prob) is not significant (p=0.56), while it significantly reduces $Δ$Prob on two of the other three models.

Related papers

• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• A Unified Model for Cross-Domain Clone Detection via Model Merging [9.757638656919832]
  Current deep learning detectors degrade significantly outside their training distribution.<n>We investigate model merging, a family of post-hoc techniques that operate solely on trained checkpoints.<n>Same-base TIES merging creates effective cross-domain detectors, validated across two model families and three random seeds.
  arXiv  Detail & Related papers  (2026-08-04T20:32:28Z)
• Distribution-First Population Simulation: Collapse, Calibration, and Recall in Non-WEIRD LLM Persona Modeling [0.0]
  Synthetic-population tools run every individual as an independent large language model (LLM) agent.<n>We show that this paradigm has a basic failure mode, and we set a distribution-first corrective against it.
  arXiv  Detail & Related papers  (2026-07-17T15:45:57Z)
• The Surprising Universality of LLM Outputs: A Real-Time Verification Primitive [0.0]
  CPU-only scoring primitive runs at 2.6 microseconds per token.<n> token rank-frequency distributions converge to the same two- parameter Mandelbrot ranking distribution.<n>We derive a single-pass scoring primitive that composes with model log probabilities when available and degrades to a rank-only mode usable on closed APIs.
  arXiv  Detail & Related papers  (2026-04-28T13:35:31Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• $V_1$: Unifying Generation and Self-Verification for Parallel Reasoners [69.66089681814013]
  $V_$ is a framework that unifies generation and verification through efficient pairwise ranking.<n>$V_$-Infer improves Pass@1 by up to $10%$ over pointwise verification.<n>$V_$-PairRL achieves $7$--$9%$ test-time scaling gains over standard RL and pointwise joint training.
  arXiv  Detail & Related papers  (2026-03-04T17:22:16Z)
• A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning [31.861874030715953]
  We provide the first theoretical framework for analyzing sampling-based test-time scaling methods.<n>We introduce RPC, a hybrid method that leverages our theoretical insights through two key components: Perplexity Consistency and Reasoning Pruning.<n>RPC achieves reasoning performance comparable to self-consistency while not only enhancing confidence reliability but also reducing sampling costs by 50%.
  arXiv  Detail & Related papers  (2025-10-17T08:59:30Z)
• Ensemble Threshold Calibration for Stable Sensitivity Control [0.0]
  We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs.<n>Our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.
  arXiv  Detail & Related papers  (2025-10-02T15:22:28Z)
• Rethinking Reward Models for Multi-Domain Test-Time Scaling [91.76069784586149]
  Prior work generally assumes that process reward models (PRMs) outperform outcome reward models (ORMs) that assess only the final answer.<n>We present the first unified evaluation of four reward model variants across 14 diverse domains.<n>We attribute this to PRM-style stepwise scoring, which inherits label noise from LLM auto-labeling and has difficulty evaluating long reasoning trajectories.
  arXiv  Detail & Related papers  (2025-10-01T04:21:14Z)
• SEAL: Steerable Reasoning Calibration of Large Language Models for Free [58.931194824519935]
  Large Language Models (LLMs) have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism.<n>Recent studies reveal substantial redundancy in the CoT reasoning traces, which negatively impacts model performance.<n>We introduce SEAL, a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains.
  arXiv  Detail & Related papers  (2025-04-07T02:42:07Z)
• Theoretically Achieving Continuous Representation of Oriented Bounding Boxes [64.15627958879053]
  This paper endeavors to completely solve the issue of discontinuity in Oriented Bounding Box representation. We propose a novel representation method called Continuous OBB (COBB) which can be readily integrated into existing detectors. For fairness and transparency of experiments, we have developed a modularized benchmark based on the open-source deep learning framework Jittor's detection toolbox JDet for OOD evaluation.
  arXiv  Detail & Related papers  (2024-02-29T09:27:40Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.