{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","short_pith_number":"pith:FHXUMIEK","canonical_record":{"source":{"id":"2605.14465","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T07:00:26Z","cross_cats_sorted":[],"title_canon_sha256":"d621ba3c98dd9262c7ecb24095d95a674207be560406ebfead348bb859db2e4a","abstract_canon_sha256":"9e7f88313850a5eaee3ab810d62ea492776f67bf233a5c15e27493754773f730"},"schema_version":"1.0"},"canonical_sha256":"29ef46208a2894c64710ed43b3ebd5913f1406d4c3ba7d4c75de612c9832ec84","source":{"kind":"arxiv","id":"2605.14465","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14465","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14465v1","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14465","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"pith_short_12","alias_value":"FHXUMIEKFCKM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"FHXUMIEKFCKMMRYQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"FHXUMIEK","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14465","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T07:00:26Z","cross_cats_sorted":[],"title_canon_sha256":"d621ba3c98dd9262c7ecb24095d95a674207be560406ebfead348bb859db2e4a","abstract_canon_sha256":"9e7f88313850a5eaee3ab810d62ea492776f67bf233a5c15e27493754773f730"},"schema_version":"1.0"},"canonical_sha256":"29ef46208a2894c64710ed43b3ebd5913f1406d4c3ba7d4c75de612c9832ec84","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:06.732158Z","signature_b64":"wn95A7Lx2xKQjVZ+/0yEctJjO53uXChKXofCBbWjdI2g3kFG5TqEins6akMw7f7HbQ1lXGp+gdeMqdqxddkaAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"29ef46208a2894c64710ed43b3ebd5913f1406d4c3ba7d4c75de612c9832ec84","last_reissued_at":"2026-05-17T23:39:06.731364Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:06.731364Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14465","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DGEjX3HxFGw+Lrbc7q2xr9yLzmYnox7nx4jSqlvgsqX5mGPetfTkrTW8nmeHmXxOgZtI5cng1JUPQiXsUwxnAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T02:08:53.837344Z"},"content_sha256":"16858fe3c72b8a497ff8895476027b85c71d64e2fb495e3ac848b5f0d0ef0af2","schema_version":"1.0","event_id":"sha256:16858fe3c72b8a497ff8895476027b85c71d64e2fb495e3ac848b5f0d0ef0af2"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"From Table to Cell: Attention for Better Reasoning with TABALIGN","license":"http://creativecommons.org/licenses/by/4.0/","headline":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chunhe Wang, Guang Cheng, Hanwei Wu, Lei Ding, Tung Sum Thomas Kwok, Xiaofeng Lin, Xinyu Wang, Zeyong Zhang, Zhijiang Guo","submitted_at":"2026-05-14T07:00:26Z","abstract_excerpt":"Multi-step LLM reasoning over structured tables fails because planning and execution share no explicit cell-grounding contract. Existing methods constrain the planner to a left-to-right factorization at odds with table permutation invariance, and score intermediate states by generated content alone, overlooking cell grounding. We conduct a pilot study showing that diffusion language models (DLMs) produce more human-aligned and permutation-stable cell attention on tables than autoregressive models, with a 40.2% median reduction in attention-AUROC variability under row reordering. Motivated by t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 1,600 human-verified attention standards used to train TABATTN are representative and stable across different table layouts, domains, and model scales; if they are not, the verifier's scoring may not reliably enforce the cell-grounding contract.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c3522d648c2513534933f5c268480c369bc6378ce1d392ac19867ded80cb822d"},"source":{"id":"2605.14465","kind":"arxiv","version":1},"verdict":{"id":"650f87fb-fe46-407e-b4bd-8ef7b50b9cb6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:25:47.957637Z","strongest_claim":"Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner.","one_line_summary":"TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 1,600 human-verified attention standards used to train TABATTN are representative and stable across different table layouts, domains, and model scales; if they are not, the verifier's scoring may not reliably enforce the cell-grounding contract.","pith_extraction_headline":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning."},"references":{"count":87,"sample":[{"doi":"","year":2025,"title":"Adapting autoregressive vision language models for parallel diffusion decoding, 2025","work_id":"7ffb24b6-5db0-46fd-a761-f5611015ec40","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg","work_id":"368943f8-485f-4c02-aa7b-601f5fc8ce36","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"URLhttps://openreview.net/forum?id=h7-XixPCAL","work_id":"54befcc6-ed61-456b-bfa5-58d9d4a3238e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","ref_index":4,"cited_arxiv_id":"2511.21631","is_internal_anchor":true},{"doi":"","year":2025,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","ref_index":5,"cited_arxiv_id":"2502.13923","is_internal_anchor":true}],"resolved_work":87,"snapshot_sha256":"bff895ffc51f7c669858c88e4a91ae9781cccab955420665c020172f61480284","internal_anchors":15},"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":"650f87fb-fe46-407e-b4bd-8ef7b50b9cb6"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"eDYxSq41MRXu+ZWNErJN5ZU1uorjA3cZzorf7JCmn2ccsVPa66QjEJ/dvE8pk9wsZO8z6R5Nuc8SffYOUPHsCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T02:08:53.838274Z"},"content_sha256":"921ae89b24640efde4b60a762f0b1a0185c1fc0ebc9f8deb37d49ca3ac5e4043","schema_version":"1.0","event_id":"sha256:921ae89b24640efde4b60a762f0b1a0185c1fc0ebc9f8deb37d49ca3ac5e4043"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2023.eacl-main.121.URLhttps://aclanthology.org/2023.eacl-main.121/) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"Kalpesh Krishna, Erin Bransom, Bailey Kuehl, Mohit Iyyer, Pradeep Dasigi, Arman Cohan, and Kyle Lo. LongEval: Guidelines for human evaluation of faithfulness in long-form summariza- tion. In Andreas Vlachos and Isabelle Augenstein, editors,","arxiv_id":"2605.14465","detector":"doi_compliance","evidence":{"ref_index":26,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"Kalpesh Krishna, Erin Bransom, Bailey Kuehl, Mohit Iyyer, Pradeep Dasigi, Arman Cohan, and Kyle Lo. LongEval: Guidelines for human evaluation of faithfulness in long-form summariza- tion. In Andreas Vlachos and Isabelle Augenstein, editors,","reconstructed_doi":"10.18653/v1/2023.eacl-main.121.URLhttps://aclanthology.org/2023.eacl-main.121/"},"severity":"advisory","ref_index":26,"audited_at":"2026-05-19T05:31:04.908303Z","event_type":"pith.integrity.v1","detected_doi":"10.18653/v1/2023.eacl-main.121.URLhttps://aclanthology.org/2023.eacl-main.121/","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"7cbd423462fcc6052ab0229d236904bb891987ff2a1560b626d3f344deb453ee","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":28,"payload_sha256":"a6110811a3c563ef1798130efdf3f1dd1b197a526ea048f72ba7f6c9f0bbbab4","signature_b64":"KZxRJftXa/MZJDm1x0hrboKI3+dpdMPBiqmAzH8GgvZzeOWiu5eB6skZ/U2I5yqmKqYdCa+i4DHfzAJpST4UBA==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T05:31:20Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4vqBXpny8+u5Bq6zVmvmE5a+NA1DjhxeqYhd3trtxO1/n6YPP+bCyGC+nPRYMUYltQabg6oGKXKS8S4MM4erDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T02:08:53.839761Z"},"content_sha256":"1e75e00b0df1ba2735017ab27b267906e3ebcb827ae4c20102e100c1d7519633","schema_version":"1.0","event_id":"sha256:1e75e00b0df1ba2735017ab27b267906e3ebcb827ae4c20102e100c1d7519633"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2024.naacl-long.26.URLhttps://aclanthology.org/2024.naacl-long.26/) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"Tianyang Liu, Fei Wang, and Muhao Chen. Rethinking tabular data understanding with large language models. In Kevin Duh, Helena Gomez, and Steven Bethard, editors,Proceedings of the 2024 Conference of the North American Chapter of the Associ","arxiv_id":"2605.14465","detector":"doi_compliance","evidence":{"ref_index":35,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"Tianyang Liu, Fei Wang, and Muhao Chen. Rethinking tabular data understanding with large language models. In Kevin Duh, Helena Gomez, and Steven Bethard, editors,Proceedings of the 2024 Conference of the North American Chapter of the Associ","reconstructed_doi":"10.18653/v1/2024.naacl-long.26.URLhttps://aclanthology.org/2024.naacl-long.26/"},"severity":"advisory","ref_index":35,"audited_at":"2026-05-19T05:31:04.908303Z","event_type":"pith.integrity.v1","detected_doi":"10.18653/v1/2024.naacl-long.26.URLhttps://aclanthology.org/2024.naacl-long.26/","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"dd6771da7f937e5b704061ea40a9ab39759ae0d07281f4c5028832863bd7e5dd","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":27,"payload_sha256":"f7dd9dd5a2e45b5b4b961bf7bffeb923e897b35b0461c37d5974df67b4f950d2","signature_b64":"0f/Mj4hOU1sQZl6lhnd0Nw6/Da4SegUPdslqCKbyfLxlt0xc3De+j5+gBAA6uRavZhtFQQf7D5z4fi6e9cQfBg==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T05:31:20Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ns8aI3guy+FkkJHab75wjxJuVbiQSd+M9FBzoGoEEQkDlcdI1LCHjYKbxP18OD3nM/G8oi0+utIC2CPy383UBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T02:08:53.840342Z"},"content_sha256":"fcd2c994dd1c758f221b516bbef9aa5eab415291b184febb64169c7de9038c07","schema_version":"1.0","event_id":"sha256:fcd2c994dd1c758f221b516bbef9aa5eab415291b184febb64169c7de9038c07"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","target":"integrity","payload":{"note":"Identifier '10.5555/3692070.3693403' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","snippet":"Aaron Lou, Chenlin Meng, and Stefano Ermon. Discrete diffusion modeling by estimating the ratios of the data distribution. InProceedings of the 41st International Conference on Machine Learning, ICML’24, 2024. URLhttps://dl.acm.org/doi/10.5","arxiv_id":"2605.14465","detector":"doi_compliance","evidence":{"doi":"10.5555/3692070.3693403","arxiv_id":null,"ref_index":37,"raw_excerpt":"Aaron Lou, Chenlin Meng, and Stefano Ermon. Discrete diffusion modeling by estimating the ratios of the data distribution. InProceedings of the 41st International Conference on Machine Learning, ICML’24, 2024. URLhttps://dl.acm.org/doi/10.5555/3692070.3693403","verdict_class":"cross_source","checked_sources":["crossref_by_doi","openalex_by_doi","doi_org_head"]},"severity":"critical","ref_index":37,"audited_at":"2026-05-19T05:31:04.908303Z","event_type":"pith.integrity.v1","detected_doi":"10.5555/3692070.3693403","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"unresolvable_identifier","evidence_hash":"c8dbb9316ca87d01f2607c65c5882c7514b9e9b425f2c8780468c2c0e9fd8cf0","paper_version":1,"verdict_class":"cross_source","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":26,"payload_sha256":"b560b05e558a49d1827178f5bd69104255e3e8f73ffb920d38be255bca11f78d","signature_b64":"ZIWK9YY9DINJxq8jN18Yb5whIs/2d/1OpJp1vKhKySYluPfkMttstOmPnWHAbj8x5DqU0WdL5YHOowZlpu8rAA==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T05:31:20Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pLE9cyFCvlYu7UoSxLBcDEW252NnoIV8t/komcSELdnNuakPLgtPZCUMFltxzgpaDYkq1giCzZM4DGxGgfnIDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T02:08:53.840945Z"},"content_sha256":"551ff17555c7ff2af5adb81e28297dd9cd32d5be767fe9dd303b1c3b6ee752c4","schema_version":"1.0","event_id":"sha256:551ff17555c7ff2af5adb81e28297dd9cd32d5be767fe9dd303b1c3b6ee752c4"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/bundle.json","state_url":"https://pith.science/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/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":"2026-05-21T02:08:53Z","links":{"resolver":"https://pith.science/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE","bundle":"https://pith.science/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/bundle.json","state":"https://pith.science/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/state.json","well_known_bundle":"https://pith.science/.well-known/pith/FHXUMIEKFCKMMRYQ5VB3H26VSE/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:FHXUMIEKFCKMMRYQ5VB3H26VSE","merge_version":"pith-open-graph-merge-v1","event_count":5,"valid_event_count":5,"invalid_event_count":0,"equivocation_count":1,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"9e7f88313850a5eaee3ab810d62ea492776f67bf233a5c15e27493754773f730","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T07:00:26Z","title_canon_sha256":"d621ba3c98dd9262c7ecb24095d95a674207be560406ebfead348bb859db2e4a"},"schema_version":"1.0","source":{"id":"2605.14465","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14465","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14465v1","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14465","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"pith_short_12","alias_value":"FHXUMIEKFCKM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"FHXUMIEKFCKMMRYQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"FHXUMIEK","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:921ae89b24640efde4b60a762f0b1a0185c1fc0ebc9f8deb37d49ca3ac5e4043","target":"graph","created_at":"2026-05-17T23:39:06Z","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":"Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The 1,600 human-verified attention standards used to train TABATTN are representative and stable across different table layouts, domains, and model scales; if they are not, the verifier's scoring may not reliably enforce the cell-grounding contract."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning."}],"snapshot_sha256":"c3522d648c2513534933f5c268480c369bc6378ce1d392ac19867ded80cb822d"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Multi-step LLM reasoning over structured tables fails because planning and execution share no explicit cell-grounding contract. Existing methods constrain the planner to a left-to-right factorization at odds with table permutation invariance, and score intermediate states by generated content alone, overlooking cell grounding. We conduct a pilot study showing that diffusion language models (DLMs) produce more human-aligned and permutation-stable cell attention on tables than autoregressive models, with a 40.2% median reduction in attention-AUROC variability under row reordering. Motivated by t","authors_text":"Chunhe Wang, Guang Cheng, Hanwei Wu, Lei Ding, Tung Sum Thomas Kwok, Xiaofeng Lin, Xinyu Wang, Zeyong Zhang, Zhijiang Guo","cross_cats":[],"headline":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T07:00:26Z","title":"From Table to Cell: Attention for Better Reasoning with TABALIGN"},"references":{"count":87,"internal_anchors":15,"resolved_work":87,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Adapting autoregressive vision language models for parallel diffusion decoding, 2025","work_id":"7ffb24b6-5db0-46fd-a761-f5611015ec40","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg","work_id":"368943f8-485f-4c02-aa7b-601f5fc8ce36","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"URLhttps://openreview.net/forum?id=h7-XixPCAL","work_id":"54befcc6-ed61-456b-bfa5-58d9d4a3238e","year":null},{"cited_arxiv_id":"2511.21631","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","year":2025},{"cited_arxiv_id":"2502.13923","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","year":2025}],"snapshot_sha256":"bff895ffc51f7c669858c88e4a91ae9781cccab955420665c020172f61480284"},"source":{"id":"2605.14465","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T01:25:47.957637Z","id":"650f87fb-fe46-407e-b4bd-8ef7b50b9cb6","model_set":{"reader":"grok-4.3"},"one_line_summary":"TABALIGN pairs a diffusion language model planner emitting binary cell masks with a trained attention verifier, raising average accuracy 15.76 points over strong baselines on eight table benchmarks while speeding execution 44.64%.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"TABALIGN pairs a diffusion language model planner that emits binary cell masks with an attention verifier trained on human standards to enforce cell grounding in table reasoning.","strongest_claim":"Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner.","weakest_assumption":"The 1,600 human-verified attention standards used to train TABATTN are representative and stable across different table layouts, domains, and model scales; if they are not, the verifier's scoring may not reliably enforce the cell-grounding contract."}},"verdict_id":"650f87fb-fe46-407e-b4bd-8ef7b50b9cb6"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:16858fe3c72b8a497ff8895476027b85c71d64e2fb495e3ac848b5f0d0ef0af2","target":"record","created_at":"2026-05-17T23:39:06Z","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":"9e7f88313850a5eaee3ab810d62ea492776f67bf233a5c15e27493754773f730","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T07:00:26Z","title_canon_sha256":"d621ba3c98dd9262c7ecb24095d95a674207be560406ebfead348bb859db2e4a"},"schema_version":"1.0","source":{"id":"2605.14465","kind":"arxiv","version":1}},"canonical_sha256":"29ef46208a2894c64710ed43b3ebd5913f1406d4c3ba7d4c75de612c9832ec84","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"29ef46208a2894c64710ed43b3ebd5913f1406d4c3ba7d4c75de612c9832ec84","first_computed_at":"2026-05-17T23:39:06.731364Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:06.731364Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"wn95A7Lx2xKQjVZ+/0yEctJjO53uXChKXofCBbWjdI2g3kFG5TqEins6akMw7f7HbQ1lXGp+gdeMqdqxddkaAQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:06.732158Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14465","source_kind":"arxiv","source_version":1}}},"equivocations":[{"signer_id":"pith.science","event_type":"integrity_finding","target":"integrity","event_ids":["sha256:1e75e00b0df1ba2735017ab27b267906e3ebcb827ae4c20102e100c1d7519633","sha256:551ff17555c7ff2af5adb81e28297dd9cd32d5be767fe9dd303b1c3b6ee752c4","sha256:fcd2c994dd1c758f221b516bbef9aa5eab415291b184febb64169c7de9038c07"]}],"invalid_events":[],"applied_event_ids":["sha256:16858fe3c72b8a497ff8895476027b85c71d64e2fb495e3ac848b5f0d0ef0af2","sha256:921ae89b24640efde4b60a762f0b1a0185c1fc0ebc9f8deb37d49ca3ac5e4043"],"state_sha256":"d5f9c64eee135b3ea75a6cd43379dd1f316852ba0f4f1e3cceeeccf51d40e6b0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HXFyxRgq8WfR06g0hXYJN/XkA4NDkF7u2FlMiilw5IobhlM4zUbBhugUkXqfqyi4BdwgQLuHqGsBwp0kt4HLAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-21T02:08:53.845873Z","bundle_sha256":"c58b5b78a0c7b830f23d9c0e3fb64c1147f4dba8ccad142b44b0aca2673202cd"}}