{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ZS7WZ6NSEOSE32GVGPJYBM7SR4","short_pith_number":"pith:ZS7WZ6NS","schema_version":"1.0","canonical_sha256":"ccbf6cf9b223a44de8d533d380b3f28f3357533f5d03dfb81c51aeb1f5fd8765","source":{"kind":"arxiv","id":"2305.10160","version":2},"attestation_state":"computed","paper":{"title":"Stop Uploading Test Data in Plain Text: Practical Strategies for Mitigating Data Contamination by Evaluation Benchmarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alon Jacovi, Avi Caciularu, Omer Goldman, Yoav Goldberg","submitted_at":"2023-05-17T12:23:38Z","abstract_excerpt":"Data contamination has become prevalent and challenging with the rise of models pretrained on large automatically-crawled corpora. For closed models, the training data becomes a trade secret, and even for open models, it is not trivial to detect contamination. Strategies such as leaderboards with hidden answers, or using test data which is guaranteed to be unseen, are expensive and become fragile with time. Assuming that all relevant actors value clean test data and will cooperate to mitigate data contamination, what can be done? We propose three strategies that can make a difference: (1) Test"},"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":"2305.10160","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-17T12:23:38Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"939bf22f924d81940f9b9bc4243b18a75abe74c83533c294a47f74968184c2dd","abstract_canon_sha256":"7d9d2902b34b76e7a3da92ee93a4921d4f762cef8f57fd0c5fb083ac62758649"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:02:01.770856Z","signature_b64":"MxWl9iNcd5+Wyr/I8aK7kFkW+q//ivroBHm8Ma1U24DbJGQ0hyDtW0JLcuUZ+Du3d5eYV6Cego8g1JFwNh/IBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ccbf6cf9b223a44de8d533d380b3f28f3357533f5d03dfb81c51aeb1f5fd8765","last_reissued_at":"2026-07-05T07:02:01.770369Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:02:01.770369Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stop Uploading Test Data in Plain Text: Practical Strategies for Mitigating Data Contamination by Evaluation Benchmarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Alon Jacovi, Avi Caciularu, Omer Goldman, Yoav Goldberg","submitted_at":"2023-05-17T12:23:38Z","abstract_excerpt":"Data contamination has become prevalent and challenging with the rise of models pretrained on large automatically-crawled corpora. For closed models, the training data becomes a trade secret, and even for open models, it is not trivial to detect contamination. Strategies such as leaderboards with hidden answers, or using test data which is guaranteed to be unseen, are expensive and become fragile with time. Assuming that all relevant actors value clean test data and will cooperate to mitigate data contamination, what can be done? We propose three strategies that can make a difference: (1) Test"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.10160","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/2305.10160/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":"2305.10160","created_at":"2026-07-05T07:02:01.770427+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.10160v2","created_at":"2026-07-05T07:02:01.770427+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.10160","created_at":"2026-07-05T07:02:01.770427+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZS7WZ6NSEOSE","created_at":"2026-07-05T07:02:01.770427+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZS7WZ6NSEOSE32GV","created_at":"2026-07-05T07:02:01.770427+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZS7WZ6NS","created_at":"2026-07-05T07:02:01.770427+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"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":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28405","citing_title":"Measuring Progress Toward AGI: A Cognitive Framework","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2507.01955","citing_title":"How Well Does GPT-4o Understand Vision? Evaluating Multimodal Foundation Models on Standard Computer Vision Tasks","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2402.13228","citing_title":"Smaug: Fixing Failure Modes of Preference Optimisation with DPO-Positive","ref_index":124,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12673","citing_title":"Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4","json":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4.json","graph_json":"https://pith.science/api/pith-number/ZS7WZ6NSEOSE32GVGPJYBM7SR4/graph.json","events_json":"https://pith.science/api/pith-number/ZS7WZ6NSEOSE32GVGPJYBM7SR4/events.json","paper":"https://pith.science/paper/ZS7WZ6NS"},"agent_actions":{"view_html":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4","download_json":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4.json","view_paper":"https://pith.science/paper/ZS7WZ6NS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.10160&json=true","fetch_graph":"https://pith.science/api/pith-number/ZS7WZ6NSEOSE32GVGPJYBM7SR4/graph.json","fetch_events":"https://pith.science/api/pith-number/ZS7WZ6NSEOSE32GVGPJYBM7SR4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4/action/storage_attestation","attest_author":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4/action/author_attestation","sign_citation":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4/action/citation_signature","submit_replication":"https://pith.science/pith/ZS7WZ6NSEOSE32GVGPJYBM7SR4/action/replication_record"}},"created_at":"2026-07-05T07:02:01.770427+00:00","updated_at":"2026-07-05T07:02:01.770427+00:00"}