LOCKR: A Hidden-State Trajectory-Guided Planner for Detecting and Repairing Stable-but-Wrong Lock-In in Diffusion Language Models

Researchers introduced LOCKR, a hidden-state trajectory-guided planner that detects and repairs "stable-but-wrong lock-in" in diffusion language models, according to the arXiv paper 2609.27220v1. Across two diffusion language models and three mathematical reasoning benchmarks, LOCKR delivered absolute accuracy gains of 2.21 to 5.37 percentage points in all five evaluated settings, with repair rates between 22% and 41%. The work reports that hidden-state trajectories outperform surface-level decoding signals such as confidence, entropy, margin, and answer stability for both detecting wrong lock-in and selecting repairs.

arXiv:2609.27220v1 Announce Type: new Abstract: Diffusion language models generate text through iterative denoising, exposing intermediate trajectories before final answers are produced. We identify a recurring reasoning failure, stable-but-wrong lock-in, where an answer stabilizes early around an incorrect value while substantial denoising remains. Surface-level decoding signals such as confidence, entropy, margin, and answer stability are insufficient to reliably distinguish correct from erroneous lock-in. We formulate selective reasoning repair as a lightweight test-time planning problem and propose LOCKR, a hidden-state trajectory-guided planner that decides when to allocate additional computation, expands a structured set of targeted repair branches, and selects the most promising continuation using trajectory-aware verification. Across two diffusion language models and three mathematical reasoning benchmarks, hidden-state trajectories consistently outperform surface signals and single hidden snapshots for both wrong-lock-in detection and repair selection. On natural evaluation distributions, LOCKR yields absolute accuracy gains of 2.21--5.37 percentage points across all five evaluated settings, with repair rates ranging from 22% to 41%. These results establish hidden diffusion trajectories as actionable signals for selective test-time reasoning repair.