{"paper":{"title":"RubricRefine: Improving Tool-Use Agent Reliability with Training-Free Pre-Execution Refinement","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"RubricRefine generates task-specific rubrics to score and repair tool-use code for contract violations before any execution occurs.","cross_cats":["cs.SE"],"primary_cat":"cs.LG","authors_text":"Abhay Venkatesh, Brendan Evers, Sam Saltwick, Will LeVine","submitted_at":"2026-05-10T19:57:32Z","abstract_excerpt":"Iterative self-refinement is a popular inference-time reliability technique, but its effectiveness in code-mode tool use depends heavily on the structure of the feedback signal: unstructured critique helps inconsistently across models, and even revision with real execution feedback improves only modestly ($0.75$ vs. $0.65$ baseline). The dominant failures are inter-tool contract violations (wrong output shape, incorrect tool routing, broken argument provenance) that run to completion without raising errors, making runtime feedback insufficient. We introduce RubricRefine, a training-free method"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"With zero execution attempts, RubricRefine reaches 0.86 on M3ToolEval averaged across seven models-improving over prior inference-time baselines on every model tested on this benchmark, at 2.6X lower latency than the strongest non-iterative alternative.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That inter-tool contract violations (wrong output shape, incorrect routing, broken argument provenance) are the dominant failures and that automatically generated rubrics can reliably detect them without any execution feedback or model-specific tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RubricRefine improves tool-use agent reliability to 0.86 on M3ToolEval by generating rubrics for pre-execution contract checking and iterative repair, outperforming baselines at 2.6X lower latency while showing no gain on single-step tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"RubricRefine generates task-specific rubrics to score and repair tool-use code for contract violations before any execution occurs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2c948d2ed6f63b8e18f797103f76dd96673eb6bad91bf6f0212ec58c7af17181"},"source":{"id":"2605.09730","kind":"arxiv","version":3},"verdict":{"id":"c81479f9-594f-4201-9856-945bb3f71783","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:21:43.945619Z","strongest_claim":"With zero execution attempts, RubricRefine reaches 0.86 on M3ToolEval averaged across seven models-improving over prior inference-time baselines on every model tested on this benchmark, at 2.6X lower latency than the strongest non-iterative alternative.","one_line_summary":"RubricRefine improves tool-use agent reliability to 0.86 on M3ToolEval by generating rubrics for pre-execution contract checking and iterative repair, outperforming baselines at 2.6X lower latency while showing no gain on single-step tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That inter-tool contract violations (wrong output shape, incorrect routing, broken argument provenance) are the dominant failures and that automatically generated rubrics can reliably detect them without any execution feedback or model-specific tuning.","pith_extraction_headline":"RubricRefine generates task-specific rubrics to score and repair tool-use code for contract violations before any execution occurs."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09730/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T16:37:11.934993Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:18.117618Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:59:29.735265Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"9fea2bff5d0949ab4a9ea3e1e4d394c8ec88be24629387903ac983ee562c7d2d"},"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"}