{"paper":{"title":"The Model Says Walk: How Surface Heuristics Override Implicit Constraints in LLM Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Surface distance cues override implicit feasibility constraints in large language models, causing systematic reasoning failures.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Lu Zhang, Ramayya Krishnan, Rema Padman, Tianchong Jiang, Yubo Li","submitted_at":"2026-03-30T21:36:09Z","abstract_excerpt":"Large language models fail when a salient surface cue conflicts with an unstated feasibility constraint. We introduce the Heuristic Override Benchmark (HOB): 500 instances spanning 4 heuristic families and 5 constraint families, with minimal pairs and explicitness gradients. We pair HOB with a falsifiable behavioral characterization following a diagnose-measure-bridge-treat arc. Causal-behavioral analysis of the car wash problem across six models reveals context-independent sigmoid heuristics: the distance cue has 8.7 to 38 times more influence than the goal, and attribution better matches key"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Causal-behavioral analysis of the car wash problem across six models reveals approximately context-independent sigmoid heuristics: the distance cue exerts 8.7 to 38 times more influence than the goal.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That minimal pairs and explicitness gradients in the HOB benchmark successfully isolate heuristic override from other factors such as knowledge gaps or prompt formatting effects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LLMs prioritize surface heuristics such as distance cues over implicit constraints in reasoning tasks, with the new HOB benchmark showing no model exceeds 75% strict accuracy and hints recovering performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Surface distance cues override implicit feasibility constraints in large language models, causing systematic reasoning failures.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f4176ba450c4737151aedc50faaac3f204cf3134f55e729b0e0cd9c01a71e615"},"source":{"id":"2603.29025","kind":"arxiv","version":3},"verdict":{"id":"5cd1875d-14d2-4a54-b211-bcd6569bbc1a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:03:13.023690Z","strongest_claim":"Causal-behavioral analysis of the car wash problem across six models reveals approximately context-independent sigmoid heuristics: the distance cue exerts 8.7 to 38 times more influence than the goal.","one_line_summary":"LLMs prioritize surface heuristics such as distance cues over implicit constraints in reasoning tasks, with the new HOB benchmark showing no model exceeds 75% strict accuracy and hints recovering performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That minimal pairs and explicitness gradients in the HOB benchmark successfully isolate heuristic override from other factors such as knowledge gaps or prompt formatting effects.","pith_extraction_headline":"Surface distance cues override implicit feasibility constraints in large language models, causing systematic reasoning failures."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.29025/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}