{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BHE5APMKC7W52H7UQWNCJYO2W3","short_pith_number":"pith:BHE5APMK","schema_version":"1.0","canonical_sha256":"09c9d03d8a17eddd1ff4859a24e1dab6c07d0192734fbe728b25afbbd19d422a","source":{"kind":"arxiv","id":"2606.07581","version":1},"attestation_state":"computed","paper":{"title":"Training-Inference Kernel Contracts: Bounding Divergence in Post-Training and Deployment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.ET"],"primary_cat":"cs.LG","authors_text":"Bruce Changlong Xu, Lan Wu","submitted_at":"2026-05-26T21:48:33Z","abstract_excerpt":"A modern post-training pipeline often writes one symbol for its policy, pi_theta, while evaluating it through two different programs: a training kernel optimized for autograd and an inference kernel optimized for low-precision, fused, dynamically batched serving. In finite precision, these kernels can induce different distributions at identical weights, with the gap concentrated on slices that aggregate benchmarks under-represent. This paper proposes kernel contracts: a contract-first framework for specifying acceptable divergence between K_train and K_inf. A contract C = (N, S, R, O, Pi) comb"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2606.07581","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-26T21:48:33Z","cross_cats_sorted":["cs.AI","cs.ET"],"title_canon_sha256":"2c2591275326161d93b4b63cf9fccf627180b1c47e8aab143e91f2bf2e96f2f4","abstract_canon_sha256":"b6bfc82e9eb74c7c2d67ed50f3c36bb54526b8acd966b72447550d042cb032b6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T00:04:43.665163Z","signature_b64":"Vv3MXZqAqVFqV+jRWRsOlbEyba+3O+9B9FodmMfe09BqzHVR2QpDIfEoYVxfYKRyptxk0J89j0oVzYhvmajbCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"09c9d03d8a17eddd1ff4859a24e1dab6c07d0192734fbe728b25afbbd19d422a","last_reissued_at":"2026-06-09T00:04:43.664500Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T00:04:43.664500Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training-Inference Kernel Contracts: Bounding Divergence in Post-Training and Deployment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.ET"],"primary_cat":"cs.LG","authors_text":"Bruce Changlong Xu, Lan Wu","submitted_at":"2026-05-26T21:48:33Z","abstract_excerpt":"A modern post-training pipeline often writes one symbol for its policy, pi_theta, while evaluating it through two different programs: a training kernel optimized for autograd and an inference kernel optimized for low-precision, fused, dynamically batched serving. In finite precision, these kernels can induce different distributions at identical weights, with the gap concentrated on slices that aggregate benchmarks under-represent. This paper proposes kernel contracts: a contract-first framework for specifying acceptable divergence between K_train and K_inf. A contract C = (N, S, R, O, Pi) comb"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.07581","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2606.07581/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2606.07581","created_at":"2026-06-09T00:04:43.664618+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.07581v1","created_at":"2026-06-09T00:04:43.664618+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.07581","created_at":"2026-06-09T00:04:43.664618+00:00"},{"alias_kind":"pith_short_12","alias_value":"BHE5APMKC7W5","created_at":"2026-06-09T00:04:43.664618+00:00"},{"alias_kind":"pith_short_16","alias_value":"BHE5APMKC7W52H7U","created_at":"2026-06-09T00:04:43.664618+00:00"},{"alias_kind":"pith_short_8","alias_value":"BHE5APMK","created_at":"2026-06-09T00:04:43.664618+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3","json":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3.json","graph_json":"https://pith.science/api/pith-number/BHE5APMKC7W52H7UQWNCJYO2W3/graph.json","events_json":"https://pith.science/api/pith-number/BHE5APMKC7W52H7UQWNCJYO2W3/events.json","paper":"https://pith.science/paper/BHE5APMK"},"agent_actions":{"view_html":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3","download_json":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3.json","view_paper":"https://pith.science/paper/BHE5APMK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.07581&json=true","fetch_graph":"https://pith.science/api/pith-number/BHE5APMKC7W52H7UQWNCJYO2W3/graph.json","fetch_events":"https://pith.science/api/pith-number/BHE5APMKC7W52H7UQWNCJYO2W3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3/action/storage_attestation","attest_author":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3/action/author_attestation","sign_citation":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3/action/citation_signature","submit_replication":"https://pith.science/pith/BHE5APMKC7W52H7UQWNCJYO2W3/action/replication_record"}},"created_at":"2026-06-09T00:04:43.664618+00:00","updated_at":"2026-06-09T00:04:43.664618+00:00"}