{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YORG4F7PTZQ43JFSC5REZGFFAG","short_pith_number":"pith:YORG4F7P","schema_version":"1.0","canonical_sha256":"c3a26e17ef9e61cda4b217624c98a501ac7e10d5f34281307b365a06327fb15f","source":{"kind":"arxiv","id":"2310.18018","version":1},"attestation_state":"computed","paper":{"title":"NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eneko Agirre, Iker Garc\\'ia-Ferrero, Jon Ander Campos, Julen Etxaniz, Oier Lopez de Lacalle, Oscar Sainz","submitted_at":"2023-10-27T09:48:29Z","abstract_excerpt":"In this position paper, we argue that the classical evaluation on Natural Language Processing (NLP) tasks using annotated benchmarks is in trouble. The worst kind of data contamination happens when a Large Language Model (LLM) is trained on the test split of a benchmark, and then evaluated in the same benchmark. The extent of the problem is unknown, as it is not straightforward to measure. Contamination causes an overestimation of the performance of a contaminated model in a target benchmark and associated task with respect to their non-contaminated counterparts. The consequences can be very h"},"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":"2310.18018","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2023-10-27T09:48:29Z","cross_cats_sorted":[],"title_canon_sha256":"fc3efaa077ff0942bc5099013ef05e8fb921ae26cd4e93564b25a82af7b49ac2","abstract_canon_sha256":"30d5877cc4857b6f49dca9c4edd80febc0c86757db25f7b21fb8ff44adf06cd5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:05:55.900059Z","signature_b64":"oUH1iZseAh1J99k1MDWYY/ikU5L2YI1VepIM3rVXyFYMFIZHLXqt7quNxjQCuQJ8ymapodzrdGmp+GKeq7uWBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c3a26e17ef9e61cda4b217624c98a501ac7e10d5f34281307b365a06327fb15f","last_reissued_at":"2026-07-05T07:05:55.899553Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:05:55.899553Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"NLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eneko Agirre, Iker Garc\\'ia-Ferrero, Jon Ander Campos, Julen Etxaniz, Oier Lopez de Lacalle, Oscar Sainz","submitted_at":"2023-10-27T09:48:29Z","abstract_excerpt":"In this position paper, we argue that the classical evaluation on Natural Language Processing (NLP) tasks using annotated benchmarks is in trouble. The worst kind of data contamination happens when a Large Language Model (LLM) is trained on the test split of a benchmark, and then evaluated in the same benchmark. The extent of the problem is unknown, as it is not straightforward to measure. Contamination causes an overestimation of the performance of a contaminated model in a target benchmark and associated task with respect to their non-contaminated counterparts. The consequences can be very h"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.18018","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/2310.18018/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":"2310.18018","created_at":"2026-07-05T07:05:55.899614+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.18018v1","created_at":"2026-07-05T07:05:55.899614+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.18018","created_at":"2026-07-05T07:05:55.899614+00:00"},{"alias_kind":"pith_short_12","alias_value":"YORG4F7PTZQ4","created_at":"2026-07-05T07:05:55.899614+00:00"},{"alias_kind":"pith_short_16","alias_value":"YORG4F7PTZQ43JFS","created_at":"2026-07-05T07:05:55.899614+00:00"},{"alias_kind":"pith_short_8","alias_value":"YORG4F7P","created_at":"2026-07-05T07:05:55.899614+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17454","citing_title":"Dissecting model behavior through agent trajectories","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01829","citing_title":"Pre-Flight: A Benchmark for Evaluating Large Language Models on Aviation Operational Knowledge","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00276","citing_title":"Testing Frontier Large Language Models' Physics Literacy in Parallel Physical Worlds","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01189","citing_title":"The Case for Model Science: Verify, Explore, Steer, Refine","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26133","citing_title":"Pretraining Data Exposure in Large Language Models: A Survey of Membership Inference, Data Contamination, and Security Implications","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24213","citing_title":"Towards Evaluation Engineering: An Empirical Study of ML Evaluation Harnesses in the Wild","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26781","citing_title":"LiveK12Bench: Have Large Multimodal Models Truly Conquered High School-level Examinations?","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23108","citing_title":"Artificial Phantasia: Emergent Mental Imagery in Large Language Models","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2507.22359","citing_title":"League of LLMs: A Benchmark-Free Paradigm for Mutual Evaluation of Large Language Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2511.02627","citing_title":"DecompSR: A dataset for decomposed analyses of compositional multihop spatial reasoning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20273","citing_title":"ActuBench: A Multi-Agent LLM Pipeline for Generation and Evaluation of Actuarial Reasoning Tasks","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18955","citing_title":"Assessing Capabilities of Large Language Models in Social Media Analytics: A Multi-task Quest","ref_index":74,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG","json":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG.json","graph_json":"https://pith.science/api/pith-number/YORG4F7PTZQ43JFSC5REZGFFAG/graph.json","events_json":"https://pith.science/api/pith-number/YORG4F7PTZQ43JFSC5REZGFFAG/events.json","paper":"https://pith.science/paper/YORG4F7P"},"agent_actions":{"view_html":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG","download_json":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG.json","view_paper":"https://pith.science/paper/YORG4F7P","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.18018&json=true","fetch_graph":"https://pith.science/api/pith-number/YORG4F7PTZQ43JFSC5REZGFFAG/graph.json","fetch_events":"https://pith.science/api/pith-number/YORG4F7PTZQ43JFSC5REZGFFAG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG/action/storage_attestation","attest_author":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG/action/author_attestation","sign_citation":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG/action/citation_signature","submit_replication":"https://pith.science/pith/YORG4F7PTZQ43JFSC5REZGFFAG/action/replication_record"}},"created_at":"2026-07-05T07:05:55.899614+00:00","updated_at":"2026-07-05T07:05:55.899614+00:00"}