{"paper":{"title":"When LLMs Stop Following Steps: A Diagnostic Study of Procedural Execution in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LLMs lose accuracy on following multi-step arithmetic procedures as length grows from 5 to 95 steps.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Abhishek Upperwal, Mayank Singh, Pritam Kadasi, Sailesh Panda","submitted_at":"2026-05-01T17:55:47Z","abstract_excerpt":"Large language models (LLMs) often achieve strong performance on reasoning benchmarks, but final-answer accuracy alone does not show whether they faithfully execute the procedure specified in a prompt. We study this question through a controlled diagnostic benchmark for procedural execution, where models are given a step-wise arithmetic algorithm and two numeric inputs, and must return the final computed value. The benchmark uses simple arithmetic operations but increases complexity through algorithm length and look-back dependencies over intermediate variables. Across 14 models and 55 dataset"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across 14 models and 55 datasets, average first-answer accuracy drops from 61% on 5-step procedures to 20% on 95-step procedures.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the measured accuracy drop reflects failures of procedural fidelity rather than unrelated factors such as context-window limits, arithmetic capability, or prompt formatting.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LLM accuracy on controlled procedural arithmetic drops from 61% at 5 steps to 20% at 95 steps, with failures including skipped steps, premature answers, and hallucinated operations.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LLMs lose accuracy on following multi-step arithmetic procedures as length grows from 5 to 95 steps.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3530dc561fd32349a47649a7afb64426ca96dddf22040cb3ec74195d0bcd8f41"},"source":{"id":"2605.00817","kind":"arxiv","version":2},"verdict":{"id":"d170fc2a-8b53-4efc-a5b0-532cb5c232c9","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T18:48:26.279158Z","strongest_claim":"Across 14 models and 55 datasets, average first-answer accuracy drops from 61% on 5-step procedures to 20% on 95-step procedures.","one_line_summary":"LLM accuracy on controlled procedural arithmetic drops from 61% at 5 steps to 20% at 95 steps, with failures including skipped steps, premature answers, and hallucinated operations.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the measured accuracy drop reflects failures of procedural fidelity rather than unrelated factors such as context-window limits, arithmetic capability, or prompt formatting.","pith_extraction_headline":"LLMs lose accuracy on following multi-step arithmetic procedures as length grows from 5 to 95 steps."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00817/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T18:41:01.458485Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T17:48:49.841220Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"e2155f906a7b2a39fc3bb17418856f5e943248c4e7b4e1c5ceb753f3797447b2"},"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"}