{"paper":{"title":"From Procedural Skills to Strategy Genes: Towards Experience-Driven Test-Time Evolution","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A compact Gene representation for reusable experience outperforms documentation-heavy Skill packages in guiding AI code solvers and enabling iterative evolution.","cross_cats":["cs.CL"],"primary_cat":"cs.SE","authors_text":"Haoyang Zhang, Junjie Wang, Yiming Ren","submitted_at":"2026-04-16T14:55:49Z","abstract_excerpt":"This beta technical report asks how reusable experience should be represented so that it can function as effective test-time control and as a substrate for iterative evolution. We study this question in 4.590 controlled trials across 45 scientific code-solving scenarios. We find that documentation-oriented Skill packages provide unstable control: their useful signal is sparse, and expanding a compact experience object into a fuller documentation package often fails to help and can degrade the overall average. We further show that representation itself is a first-order factor. A compact Gene re"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"A compact Gene representation yields the strongest overall average, remains competitive under substantial structural perturbations, and outperforms matched-budget Skill fragments, while reattaching documentation-oriented material usually weakens rather than improves it. On CritPt, gene-evolved systems improve over their paired base models from 9.1% to 18.57% and from 17.7% to 27.14%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 45 scientific code-solving scenarios and the specific definition of Gene versus Skill fragments are representative of broader experience-reuse needs, and that observed gains stem primarily from representation format rather than unstated differences in prompt construction or model behavior.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Compact Gene representations of experience outperform documentation-oriented Skill packages for test-time control and iterative evolution in code-solving tasks, with measured gains on CritPt from 9.1% to 18.57% and 17.7% to 27.14%.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A compact Gene representation for reusable experience outperforms documentation-heavy Skill packages in guiding AI code solvers and enabling iterative evolution.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a1ce6020e16cae731ef4bd7fe0f790a3a2e6c14da5babf21186f3c05c0b595f9"},"source":{"id":"2604.15097","kind":"arxiv","version":2},"verdict":{"id":"cf4b4b39-01ed-4fe5-a58b-9e8389c13310","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T10:50:00.857462Z","strongest_claim":"A compact Gene representation yields the strongest overall average, remains competitive under substantial structural perturbations, and outperforms matched-budget Skill fragments, while reattaching documentation-oriented material usually weakens rather than improves it. On CritPt, gene-evolved systems improve over their paired base models from 9.1% to 18.57% and from 17.7% to 27.14%.","one_line_summary":"Compact Gene representations of experience outperform documentation-oriented Skill packages for test-time control and iterative evolution in code-solving tasks, with measured gains on CritPt from 9.1% to 18.57% and 17.7% to 27.14%.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 45 scientific code-solving scenarios and the specific definition of Gene versus Skill fragments are representative of broader experience-reuse needs, and that observed gains stem primarily from representation format rather than unstated differences in prompt construction or model behavior.","pith_extraction_headline":"A compact Gene representation for reusable experience outperforms documentation-heavy Skill packages in guiding AI code solvers and enabling iterative evolution."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.15097/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"}