{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:HFWN3N4JEKNV24M37WMU2O533J","short_pith_number":"pith:HFWN3N4J","canonical_record":{"source":{"id":"2604.18966","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-21T01:29:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"44e1de1f9195b688031c553ed9b9ffaa2228f7764f832295c544a7c6f9d31fc8","abstract_canon_sha256":"554b1c38bbb533c41d582b88baf275014c59f2107862a135d563bcbc73cd004e"},"schema_version":"1.0"},"canonical_sha256":"396cddb789229b5d719bfd994d3bbbda717ba0be4ff3e29e140fbddc1112560e","source":{"kind":"arxiv","id":"2604.18966","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.18966","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"arxiv_version","alias_value":"2604.18966v2","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18966","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_12","alias_value":"HFWN3N4JEKNV","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_16","alias_value":"HFWN3N4JEKNV24M3","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_8","alias_value":"HFWN3N4J","created_at":"2026-05-20T00:03:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:HFWN3N4JEKNV24M37WMU2O533J","target":"record","payload":{"canonical_record":{"source":{"id":"2604.18966","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-21T01:29:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"44e1de1f9195b688031c553ed9b9ffaa2228f7764f832295c544a7c6f9d31fc8","abstract_canon_sha256":"554b1c38bbb533c41d582b88baf275014c59f2107862a135d563bcbc73cd004e"},"schema_version":"1.0"},"canonical_sha256":"396cddb789229b5d719bfd994d3bbbda717ba0be4ff3e29e140fbddc1112560e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:11.852042Z","signature_b64":"+mUPkjDp71CYIodvelpq5rQdQQI0pvoNpBeArF9u3WHVnw+zZaMmJe3IcCgSurDgZRAb8zbJqGpRJl+HmMW4Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"396cddb789229b5d719bfd994d3bbbda717ba0be4ff3e29e140fbddc1112560e","last_reissued_at":"2026-05-20T00:03:11.851100Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:11.851100Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.18966","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:03:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4OotA/dhEaJBLdKcornKGHWpIJE5Bbx0+0i3SBVqs2v5NICNmdkW+OUMh9/RKNcJcMC4XPQNwViPcwm/VQLxCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T10:52:41.524634Z"},"content_sha256":"3bbe8023316a0f1ee2ecc6a3dfc8978d73d67db1260b022021696afc61efb752","schema_version":"1.0","event_id":"sha256:3bbe8023316a0f1ee2ecc6a3dfc8978d73d67db1260b022021696afc61efb752"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:HFWN3N4JEKNV24M37WMU2O533J","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Self-Improving Tabular Language Models via Iterative Reward-Guided Post-Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexandra Brintrup, Guangya Hao, Mario Fritz, Tejumade Afonja, Yunbo Long","submitted_at":"2026-04-21T01:29:52Z","abstract_excerpt":"Tabular language models can generate synthetic tables by modeling rows as token sequences, but they are typically trained once with supervised fine-tuning and then used as static synthesizers. This is limiting because next-token likelihood does not directly optimize the distributional, utility, and indistinguishability properties used to evaluate synthetic data. We study iterative reward-guided post-training for tabular language models through a generate--score--align protocol, where a generator samples synthetic rows, a task-specified reward ranks them, and the model is updated relative to a "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show TabGRAA outperforms existing methods in fidelity, utility, and privacy, while matching or exceeding diffusion-based synthesizers, advancing tabular synthesis from static statistical replication to dynamic, self-improving generation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen automated quality signal (such as a two-sample distinguishability classifier or distance-based reward) can reliably and unbiasedly partition newly generated samples into high- and low-quality groups at each iteration without introducing artifacts or circular feedback.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TabGRAA enables self-improving tabular language models through iterative group-relative advantage alignment using modular automated quality signals like distinguishability classifiers.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3961b49fff7abc2a55aa92b2378a32f1eceabff2009dd8f3af336066ec3d44ae"},"source":{"id":"2604.18966","kind":"arxiv","version":2},"verdict":{"id":"636ccbe8-5b85-4675-94a0-7d39ef180d1d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T03:01:14.803598Z","strongest_claim":"Experiments show TabGRAA outperforms existing methods in fidelity, utility, and privacy, while matching or exceeding diffusion-based synthesizers, advancing tabular synthesis from static statistical replication to dynamic, self-improving generation.","one_line_summary":"TabGRAA enables self-improving tabular language models through iterative group-relative advantage alignment using modular automated quality signals like distinguishability classifiers.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen automated quality signal (such as a two-sample distinguishability classifier or distance-based reward) can reliably and unbiasedly partition newly generated samples into high- and low-quality groups at each iteration without introducing artifacts or circular feedback.","pith_extraction_headline":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.18966/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"},"verdict_id":"636ccbe8-5b85-4675-94a0-7d39ef180d1d"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:03:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"o1htcVzDGXfNG8FKpX9hOpzbvy5W+AM2l5TOsPTFlR8vzzsfL320oLHOX2o/XN4rByW8iEGR+cHIfDi7XGLmAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T10:52:41.525556Z"},"content_sha256":"6a3839c5a2732f0bec46fdb716ab64831f5dd6c491d602cfcdc8a2284bf2ec5c","schema_version":"1.0","event_id":"sha256:6a3839c5a2732f0bec46fdb716ab64831f5dd6c491d602cfcdc8a2284bf2ec5c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HFWN3N4JEKNV24M37WMU2O533J/bundle.json","state_url":"https://pith.science/pith/HFWN3N4JEKNV24M37WMU2O533J/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HFWN3N4JEKNV24M37WMU2O533J/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-09T10:52:41Z","links":{"resolver":"https://pith.science/pith/HFWN3N4JEKNV24M37WMU2O533J","bundle":"https://pith.science/pith/HFWN3N4JEKNV24M37WMU2O533J/bundle.json","state":"https://pith.science/pith/HFWN3N4JEKNV24M37WMU2O533J/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HFWN3N4JEKNV24M37WMU2O533J/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:HFWN3N4JEKNV24M37WMU2O533J","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"554b1c38bbb533c41d582b88baf275014c59f2107862a135d563bcbc73cd004e","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-21T01:29:52Z","title_canon_sha256":"44e1de1f9195b688031c553ed9b9ffaa2228f7764f832295c544a7c6f9d31fc8"},"schema_version":"1.0","source":{"id":"2604.18966","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.18966","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"arxiv_version","alias_value":"2604.18966v2","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18966","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_12","alias_value":"HFWN3N4JEKNV","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_16","alias_value":"HFWN3N4JEKNV24M3","created_at":"2026-05-20T00:03:11Z"},{"alias_kind":"pith_short_8","alias_value":"HFWN3N4J","created_at":"2026-05-20T00:03:11Z"}],"graph_snapshots":[{"event_id":"sha256:6a3839c5a2732f0bec46fdb716ab64831f5dd6c491d602cfcdc8a2284bf2ec5c","target":"graph","created_at":"2026-05-20T00:03:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Experiments show TabGRAA outperforms existing methods in fidelity, utility, and privacy, while matching or exceeding diffusion-based synthesizers, advancing tabular synthesis from static statistical replication to dynamic, self-improving generation."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the chosen automated quality signal (such as a two-sample distinguishability classifier or distance-based reward) can reliably and unbiasedly partition newly generated samples into high- and low-quality groups at each iteration without introducing artifacts or circular feedback."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"TabGRAA enables self-improving tabular language models through iterative group-relative advantage alignment using modular automated quality signals like distinguishability classifiers."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences."}],"snapshot_sha256":"3961b49fff7abc2a55aa92b2378a32f1eceabff2009dd8f3af336066ec3d44ae"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.18966/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Tabular language models can generate synthetic tables by modeling rows as token sequences, but they are typically trained once with supervised fine-tuning and then used as static synthesizers. This is limiting because next-token likelihood does not directly optimize the distributional, utility, and indistinguishability properties used to evaluate synthetic data. We study iterative reward-guided post-training for tabular language models through a generate--score--align protocol, where a generator samples synthetic rows, a task-specified reward ranks them, and the model is updated relative to a ","authors_text":"Alexandra Brintrup, Guangya Hao, Mario Fritz, Tejumade Afonja, Yunbo Long","cross_cats":["cs.AI"],"headline":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-21T01:29:52Z","title":"Self-Improving Tabular Language Models via Iterative Reward-Guided Post-Training"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.18966","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T03:01:14.803598Z","id":"636ccbe8-5b85-4675-94a0-7d39ef180d1d","model_set":{"reader":"grok-4.3"},"one_line_summary":"TabGRAA enables self-improving tabular language models through iterative group-relative advantage alignment using modular automated quality signals like distinguishability classifiers.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Tabular language models can self-improve by iteratively partitioning their own generations into quality groups and aligning on the differences.","strongest_claim":"Experiments show TabGRAA outperforms existing methods in fidelity, utility, and privacy, while matching or exceeding diffusion-based synthesizers, advancing tabular synthesis from static statistical replication to dynamic, self-improving generation.","weakest_assumption":"That the chosen automated quality signal (such as a two-sample distinguishability classifier or distance-based reward) can reliably and unbiasedly partition newly generated samples into high- and low-quality groups at each iteration without introducing artifacts or circular feedback."}},"verdict_id":"636ccbe8-5b85-4675-94a0-7d39ef180d1d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3bbe8023316a0f1ee2ecc6a3dfc8978d73d67db1260b022021696afc61efb752","target":"record","created_at":"2026-05-20T00:03:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"554b1c38bbb533c41d582b88baf275014c59f2107862a135d563bcbc73cd004e","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-21T01:29:52Z","title_canon_sha256":"44e1de1f9195b688031c553ed9b9ffaa2228f7764f832295c544a7c6f9d31fc8"},"schema_version":"1.0","source":{"id":"2604.18966","kind":"arxiv","version":2}},"canonical_sha256":"396cddb789229b5d719bfd994d3bbbda717ba0be4ff3e29e140fbddc1112560e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"396cddb789229b5d719bfd994d3bbbda717ba0be4ff3e29e140fbddc1112560e","first_computed_at":"2026-05-20T00:03:11.851100Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:03:11.851100Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"+mUPkjDp71CYIodvelpq5rQdQQI0pvoNpBeArF9u3WHVnw+zZaMmJe3IcCgSurDgZRAb8zbJqGpRJl+HmMW4Dg==","signature_status":"signed_v1","signed_at":"2026-05-20T00:03:11.852042Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.18966","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3bbe8023316a0f1ee2ecc6a3dfc8978d73d67db1260b022021696afc61efb752","sha256:6a3839c5a2732f0bec46fdb716ab64831f5dd6c491d602cfcdc8a2284bf2ec5c"],"state_sha256":"a903a39bf020745b647a88d8d82344bbbf40eb7522323635c6271997693ba80d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"l91XZXQPZhZo0Xn+CLUiEJlpS2ITXY4TWVAcq5qD3hPTPzo1KYu903hmbwGYXxV7y0aAXjdSEmpHXt9yoNfhDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-09T10:52:41.529918Z","bundle_sha256":"5dd3eb5a69eab02869cab62247957f029ca0b83fa12b6bdf04542b134be83da6"}}