{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MJL4FY3PVNTTNXB4IUMQ4VPDII","short_pith_number":"pith:MJL4FY3P","schema_version":"1.0","canonical_sha256":"6257c2e36fab6736dc3c45190e55e3421f075dbc311145b4a496ee78049791eb","source":{"kind":"arxiv","id":"2602.04031","version":2},"attestation_state":"computed","paper":{"title":"The Illusion of Generalization in Tabular Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aditya Gorla, Ratish Puduppully","submitted_at":"2026-02-03T21:41:30Z","abstract_excerpt":"Tabular Language Models (TLMs) have been claimed to achieve strong generalization for tabular prediction. We conduct a systematic re-evaluation of Tabula-8B as a representative TLM, utilizing 165 datasets from the UniPredict benchmark. Our investigation reveals three findings. First, binary and categorical classification achieve near-zero median lift over majority-class baselines and strong aggregate performance is driven entirely by quartile classification tasks. Second, top-performing datasets exhibit pervasive contamination, including complete train-test overlap and task-level leakage that "},"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":"2602.04031","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T21:41:30Z","cross_cats_sorted":[],"title_canon_sha256":"10a775ea978faa43bfe2b7638de15373cdd197f6ea73ac197d1e8c517c4600f7","abstract_canon_sha256":"abec0036db7a49eb0996e3292ca22a29a50b5ef9ac35931079cc74bc4839dc48"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T01:02:32.158051Z","signature_b64":"q1P5gF+/6LgY2+dPzqjCOvKlOvXmlpDbA1VsaZxNqyU3ChJqIT4BDs9AuuY8D943dAu8FZEtHhO6kWekbrhtAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6257c2e36fab6736dc3c45190e55e3421f075dbc311145b4a496ee78049791eb","last_reissued_at":"2026-06-01T01:02:32.156897Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T01:02:32.156897Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Illusion of Generalization in Tabular Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aditya Gorla, Ratish Puduppully","submitted_at":"2026-02-03T21:41:30Z","abstract_excerpt":"Tabular Language Models (TLMs) have been claimed to achieve strong generalization for tabular prediction. We conduct a systematic re-evaluation of Tabula-8B as a representative TLM, utilizing 165 datasets from the UniPredict benchmark. Our investigation reveals three findings. First, binary and categorical classification achieve near-zero median lift over majority-class baselines and strong aggregate performance is driven entirely by quartile classification tasks. Second, top-performing datasets exhibit pervasive contamination, including complete train-test overlap and task-level leakage that "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.04031","kind":"arxiv","version":2},"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/2602.04031/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":"2602.04031","created_at":"2026-06-01T01:02:32.157058+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.04031v2","created_at":"2026-06-01T01:02:32.157058+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.04031","created_at":"2026-06-01T01:02:32.157058+00:00"},{"alias_kind":"pith_short_12","alias_value":"MJL4FY3PVNTT","created_at":"2026-06-01T01:02:32.157058+00:00"},{"alias_kind":"pith_short_16","alias_value":"MJL4FY3PVNTTNXB4","created_at":"2026-06-01T01:02:32.157058+00:00"},{"alias_kind":"pith_short_8","alias_value":"MJL4FY3P","created_at":"2026-06-01T01:02:32.157058+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.12292","citing_title":"STRABLE: Benchmarking Tabular Machine Learning with Strings","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10616","citing_title":"MulTaBench: Benchmarking Multimodal Tabular Learning with Text and Image","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21696","citing_title":"Towards Universal Tabular Embeddings: A Benchmark Across Data Tasks","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII","json":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII.json","graph_json":"https://pith.science/api/pith-number/MJL4FY3PVNTTNXB4IUMQ4VPDII/graph.json","events_json":"https://pith.science/api/pith-number/MJL4FY3PVNTTNXB4IUMQ4VPDII/events.json","paper":"https://pith.science/paper/MJL4FY3P"},"agent_actions":{"view_html":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII","download_json":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII.json","view_paper":"https://pith.science/paper/MJL4FY3P","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.04031&json=true","fetch_graph":"https://pith.science/api/pith-number/MJL4FY3PVNTTNXB4IUMQ4VPDII/graph.json","fetch_events":"https://pith.science/api/pith-number/MJL4FY3PVNTTNXB4IUMQ4VPDII/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII/action/storage_attestation","attest_author":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII/action/author_attestation","sign_citation":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII/action/citation_signature","submit_replication":"https://pith.science/pith/MJL4FY3PVNTTNXB4IUMQ4VPDII/action/replication_record"}},"created_at":"2026-06-01T01:02:32.157058+00:00","updated_at":"2026-06-01T01:02:32.157058+00:00"}