{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GTLGACFW7ZNJCJPOCMBDVVJPNO","short_pith_number":"pith:GTLGACFW","schema_version":"1.0","canonical_sha256":"34d66008b6fe5a9125ee13023ad52f6bbfbe92d33d2ca17c3bb9fe949e79fc06","source":{"kind":"arxiv","id":"2608.00004","version":1},"attestation_state":"computed","paper":{"title":"Cost-Effective Automated Judging of Natural-Language Mathematical Proofs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Benjamin Grayzel","submitted_at":"2026-05-29T17:22:57Z","abstract_excerpt":"Grading natural-language mathematical proofs is a recurring cost in evaluating math-reasoning systems, and frontier LLM judges are expensive. We ask whether cheap open-weight models can serve as reliable judges given a candidate proof, a ground-truth proof, and a human-grading rubric. On a 200-instance validation sample of IMO-GradingBench, three cheap judges (GPT-OSS 120B, DeepSeek-V4 Flash, Gemma-4 31B) agree with human pass/fail decisions at rates statistically indistinguishable from Claude Opus 4.7 and Gemini 3.1 Pro, at up to $100\\times$ lower cost. We had expected a majority vote of the "},"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":"2608.00004","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-29T17:22:57Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"4bb0bd306859c556946dd0370cadb2a45c883a9c8b880304417ed3dc75c722f8","abstract_canon_sha256":"01134f8409a8c2d5a9fb66d58573d84e04a73cdf401450c45f1ad5038a738ee0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T00:31:37.271978Z","signature_b64":"mAgrkYrAzOdovDX1PaRZJRlUhmO6a8WxyUqpK0gl7Tl5LJanx8LEFsr+4JdYDRoWVonsPum0tn1h+uxqKWoNAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"34d66008b6fe5a9125ee13023ad52f6bbfbe92d33d2ca17c3bb9fe949e79fc06","last_reissued_at":"2026-08-04T00:31:37.270384Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T00:31:37.270384Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Cost-Effective Automated Judging of Natural-Language Mathematical Proofs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Benjamin Grayzel","submitted_at":"2026-05-29T17:22:57Z","abstract_excerpt":"Grading natural-language mathematical proofs is a recurring cost in evaluating math-reasoning systems, and frontier LLM judges are expensive. We ask whether cheap open-weight models can serve as reliable judges given a candidate proof, a ground-truth proof, and a human-grading rubric. On a 200-instance validation sample of IMO-GradingBench, three cheap judges (GPT-OSS 120B, DeepSeek-V4 Flash, Gemma-4 31B) agree with human pass/fail decisions at rates statistically indistinguishable from Claude Opus 4.7 and Gemini 3.1 Pro, at up to $100\\times$ lower cost. We had expected a majority vote of the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.00004","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/2608.00004/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":"2608.00004","created_at":"2026-08-04T00:31:37.271444+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.00004v1","created_at":"2026-08-04T00:31:37.271444+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.00004","created_at":"2026-08-04T00:31:37.271444+00:00"},{"alias_kind":"pith_short_12","alias_value":"GTLGACFW7ZNJ","created_at":"2026-08-04T00:31:37.271444+00:00"},{"alias_kind":"pith_short_16","alias_value":"GTLGACFW7ZNJCJPO","created_at":"2026-08-04T00:31:37.271444+00:00"},{"alias_kind":"pith_short_8","alias_value":"GTLGACFW","created_at":"2026-08-04T00:31:37.271444+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO","json":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO.json","graph_json":"https://pith.science/api/pith-number/GTLGACFW7ZNJCJPOCMBDVVJPNO/graph.json","events_json":"https://pith.science/api/pith-number/GTLGACFW7ZNJCJPOCMBDVVJPNO/events.json","paper":"https://pith.science/paper/GTLGACFW"},"agent_actions":{"view_html":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO","download_json":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO.json","view_paper":"https://pith.science/paper/GTLGACFW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.00004&json=true","fetch_graph":"https://pith.science/api/pith-number/GTLGACFW7ZNJCJPOCMBDVVJPNO/graph.json","fetch_events":"https://pith.science/api/pith-number/GTLGACFW7ZNJCJPOCMBDVVJPNO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO/action/storage_attestation","attest_author":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO/action/author_attestation","sign_citation":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO/action/citation_signature","submit_replication":"https://pith.science/pith/GTLGACFW7ZNJCJPOCMBDVVJPNO/action/replication_record"}},"created_at":"2026-08-04T00:31:37.271444+00:00","updated_at":"2026-08-04T00:31:37.271444+00:00"}