{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZGF43COKPR5TWX4VWWFVPVQPIM","short_pith_number":"pith:ZGF43COK","schema_version":"1.0","canonical_sha256":"c98bcd89ca7c7b3b5f95b58b57d60f4304d9e1ee7513cb417e973b0b060c365f","source":{"kind":"arxiv","id":"2606.26185","version":1},"attestation_state":"computed","paper":{"title":"Necessary but Not Sufficient: Temperature Control and Reproducibility in LLM-as-Judge Safety Evaluations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hiroki Tamba","submitted_at":"2026-06-24T13:24:40Z","abstract_excerpt":"LLM-as-judge (\"grader\") components are now standard in evaluation harnesses, including safety evaluations where a pass/fail verdict may gate downstream deployment decisions. A widespread assumption is that setting the grader's sampling temperature to 0 makes grading deterministic. We test this assumption against a real safety-evaluation codebase (Japan AISI's open-source aisev) and show it fails on two levels. First, the harness invokes its grader without setting temperature or seed; the underlying provider silently applies its default of 1.0, so items near the decision boundary flip pass/fail"},"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":"2606.26185","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-24T13:24:40Z","cross_cats_sorted":[],"title_canon_sha256":"a6e2d0396548ba46df2fab06478853afa230813cb0be6076e253a82ff7acca9a","abstract_canon_sha256":"45a2cbc03f7e39fb1926797c00ef67fe93d51e3f9db97b648af80b8c804756b2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T00:15:28.804230Z","signature_b64":"ydwBTenCAx84ILIxBJHQVDV22f617caw2OhrOGfYgufsB5enbyBZR8bvM/SU00YRD7W/jr/sQS1PcD1E7LCkAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c98bcd89ca7c7b3b5f95b58b57d60f4304d9e1ee7513cb417e973b0b060c365f","last_reissued_at":"2026-06-26T00:15:28.803819Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T00:15:28.803819Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Necessary but Not Sufficient: Temperature Control and Reproducibility in LLM-as-Judge Safety Evaluations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hiroki Tamba","submitted_at":"2026-06-24T13:24:40Z","abstract_excerpt":"LLM-as-judge (\"grader\") components are now standard in evaluation harnesses, including safety evaluations where a pass/fail verdict may gate downstream deployment decisions. A widespread assumption is that setting the grader's sampling temperature to 0 makes grading deterministic. We test this assumption against a real safety-evaluation codebase (Japan AISI's open-source aisev) and show it fails on two levels. First, the harness invokes its grader without setting temperature or seed; the underlying provider silently applies its default of 1.0, so items near the decision boundary flip pass/fail"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.26185","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/2606.26185/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":"2606.26185","created_at":"2026-06-26T00:15:28.803882+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.26185v1","created_at":"2026-06-26T00:15:28.803882+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.26185","created_at":"2026-06-26T00:15:28.803882+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZGF43COKPR5T","created_at":"2026-06-26T00:15:28.803882+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZGF43COKPR5TWX4V","created_at":"2026-06-26T00:15:28.803882+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZGF43COK","created_at":"2026-06-26T00:15:28.803882+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.01544","citing_title":"Measuring Product Quality Using Images: The CLIP Q-Score and an Application to Real Estate","ref_index":77,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM","json":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM.json","graph_json":"https://pith.science/api/pith-number/ZGF43COKPR5TWX4VWWFVPVQPIM/graph.json","events_json":"https://pith.science/api/pith-number/ZGF43COKPR5TWX4VWWFVPVQPIM/events.json","paper":"https://pith.science/paper/ZGF43COK"},"agent_actions":{"view_html":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM","download_json":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM.json","view_paper":"https://pith.science/paper/ZGF43COK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.26185&json=true","fetch_graph":"https://pith.science/api/pith-number/ZGF43COKPR5TWX4VWWFVPVQPIM/graph.json","fetch_events":"https://pith.science/api/pith-number/ZGF43COKPR5TWX4VWWFVPVQPIM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM/action/storage_attestation","attest_author":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM/action/author_attestation","sign_citation":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM/action/citation_signature","submit_replication":"https://pith.science/pith/ZGF43COKPR5TWX4VWWFVPVQPIM/action/replication_record"}},"created_at":"2026-06-26T00:15:28.803882+00:00","updated_at":"2026-06-26T00:15:28.803882+00:00"}