{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:J5Z7TJHM5KIX5WRPBMORWQGOIH","short_pith_number":"pith:J5Z7TJHM","schema_version":"1.0","canonical_sha256":"4f73f9a4ecea917eda2f0b1d1b40ce41d295d63b197c7cdf0d61730a40f57898","source":{"kind":"arxiv","id":"2605.26789","version":1},"attestation_state":"computed","paper":{"title":"Composition Collapse: Stable Factual Knowledge Does Not Imply Compositional Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Hongzhi Wang, Jie Chen, Meng Han, Wenpeng Xing, Xuyang Teng, Yunzhao Wei, Zhe Yu","submitted_at":"2026-05-26T10:01:01Z","abstract_excerpt":"Post-training is routinely evaluated through aggregate benchmark scores that treat multi-hop reasoning as a single capability -- as if a model that answers more questions correctly must be better at assembling facts. We show that this assumption can be misleading: recipes with statistically indistinguishable atomic knowledge produce composition behaviour separated by over 40 percentage points, a phenomenon we call composition collapse: the systematic failure to assemble stably-known facts into chains, invisible to aggregate metrics. We introduce a double-gate protocol that changes the estimand"},"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":"2605.26789","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-26T10:01:01Z","cross_cats_sorted":[],"title_canon_sha256":"9785614c733db9b39f3562149d4d5b4a56442df162a08fc3d15298fdee1cfa80","abstract_canon_sha256":"c08badc7933d2977c4be666392ee74cc2279cc6d3d51ee8366bd1de4c2fcad38"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:06:12.773199Z","signature_b64":"ZM27+GnSbbKomApab6399eaLFmeMoL1LVLiLVW4Sl5wvDiID/iaQsxlMKMtZC7C71TIj6CgpTfCp2rSYqnFhBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4f73f9a4ecea917eda2f0b1d1b40ce41d295d63b197c7cdf0d61730a40f57898","last_reissued_at":"2026-05-27T01:06:12.772137Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:06:12.772137Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Composition Collapse: Stable Factual Knowledge Does Not Imply Compositional Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Hongzhi Wang, Jie Chen, Meng Han, Wenpeng Xing, Xuyang Teng, Yunzhao Wei, Zhe Yu","submitted_at":"2026-05-26T10:01:01Z","abstract_excerpt":"Post-training is routinely evaluated through aggregate benchmark scores that treat multi-hop reasoning as a single capability -- as if a model that answers more questions correctly must be better at assembling facts. We show that this assumption can be misleading: recipes with statistically indistinguishable atomic knowledge produce composition behaviour separated by over 40 percentage points, a phenomenon we call composition collapse: the systematic failure to assemble stably-known facts into chains, invisible to aggregate metrics. We introduce a double-gate protocol that changes the estimand"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26789","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/2605.26789/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":"2605.26789","created_at":"2026-05-27T01:06:12.772257+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26789v1","created_at":"2026-05-27T01:06:12.772257+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26789","created_at":"2026-05-27T01:06:12.772257+00:00"},{"alias_kind":"pith_short_12","alias_value":"J5Z7TJHM5KIX","created_at":"2026-05-27T01:06:12.772257+00:00"},{"alias_kind":"pith_short_16","alias_value":"J5Z7TJHM5KIX5WRP","created_at":"2026-05-27T01:06:12.772257+00:00"},{"alias_kind":"pith_short_8","alias_value":"J5Z7TJHM","created_at":"2026-05-27T01:06:12.772257+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/J5Z7TJHM5KIX5WRPBMORWQGOIH","json":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH.json","graph_json":"https://pith.science/api/pith-number/J5Z7TJHM5KIX5WRPBMORWQGOIH/graph.json","events_json":"https://pith.science/api/pith-number/J5Z7TJHM5KIX5WRPBMORWQGOIH/events.json","paper":"https://pith.science/paper/J5Z7TJHM"},"agent_actions":{"view_html":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH","download_json":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH.json","view_paper":"https://pith.science/paper/J5Z7TJHM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26789&json=true","fetch_graph":"https://pith.science/api/pith-number/J5Z7TJHM5KIX5WRPBMORWQGOIH/graph.json","fetch_events":"https://pith.science/api/pith-number/J5Z7TJHM5KIX5WRPBMORWQGOIH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH/action/storage_attestation","attest_author":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH/action/author_attestation","sign_citation":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH/action/citation_signature","submit_replication":"https://pith.science/pith/J5Z7TJHM5KIX5WRPBMORWQGOIH/action/replication_record"}},"created_at":"2026-05-27T01:06:12.772257+00:00","updated_at":"2026-05-27T01:06:12.772257+00:00"}