{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TMIDWGUGVKZ6ACPECZTP6JFTDY","short_pith_number":"pith:TMIDWGUG","schema_version":"1.0","canonical_sha256":"9b103b1a86aab3e009e41666ff24b31e20085be3eac1ef7fb5df1b84e59cd726","source":{"kind":"arxiv","id":"2505.22169","version":2},"attestation_state":"computed","paper":{"title":"ReliableEval: A Recipe for Stochastic LLM Evaluation via Method of Moments","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avi Caciularu, Eliya Habba, Gabriel Stanovsky, Gili Lior, Shahar Levy","submitted_at":"2025-05-28T09:40:48Z","abstract_excerpt":"LLMs are highly sensitive to prompt phrasing, yet standard benchmarks typically report performance using a single prompt, raising concerns about the reliability of such evaluations. In this work, we argue for a stochastic method of moments evaluation over the space of meaning-preserving prompt perturbations. We introduce a formal definition of reliable evaluation that accounts for prompt sensitivity, and suggest ReliableEval - a method for estimating the number of prompt resamplings needed to obtain meaningful results. Using our framework, we stochastically evaluate five frontier LLMs and find"},"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":"2505.22169","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-28T09:40:48Z","cross_cats_sorted":[],"title_canon_sha256":"d4b5c1c1fb91eb391151c2ad7ede7a70d704a34a396b7fa31f9fd96858680b73","abstract_canon_sha256":"053e8b810ca3e96917379290f55d9fb7faad0e5c8a50020ae290d7ea46002775"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:09:09.408106Z","signature_b64":"jMRCXj10REY6S5wwaLm+hPwd/lNNN2Q6kZLsJ1AhIKePKBVsLF0H0iFzuO4DmaYDJMPsHXD+KjOfE/FNb0T3Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9b103b1a86aab3e009e41666ff24b31e20085be3eac1ef7fb5df1b84e59cd726","last_reissued_at":"2026-06-12T01:09:09.407191Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:09:09.407191Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ReliableEval: A Recipe for Stochastic LLM Evaluation via Method of Moments","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avi Caciularu, Eliya Habba, Gabriel Stanovsky, Gili Lior, Shahar Levy","submitted_at":"2025-05-28T09:40:48Z","abstract_excerpt":"LLMs are highly sensitive to prompt phrasing, yet standard benchmarks typically report performance using a single prompt, raising concerns about the reliability of such evaluations. In this work, we argue for a stochastic method of moments evaluation over the space of meaning-preserving prompt perturbations. We introduce a formal definition of reliable evaluation that accounts for prompt sensitivity, and suggest ReliableEval - a method for estimating the number of prompt resamplings needed to obtain meaningful results. Using our framework, we stochastically evaluate five frontier LLMs and find"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.22169","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/2505.22169/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":"2505.22169","created_at":"2026-06-12T01:09:09.407305+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22169v2","created_at":"2026-06-12T01:09:09.407305+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22169","created_at":"2026-06-12T01:09:09.407305+00:00"},{"alias_kind":"pith_short_12","alias_value":"TMIDWGUGVKZ6","created_at":"2026-06-12T01:09:09.407305+00:00"},{"alias_kind":"pith_short_16","alias_value":"TMIDWGUGVKZ6ACPE","created_at":"2026-06-12T01:09:09.407305+00:00"},{"alias_kind":"pith_short_8","alias_value":"TMIDWGUG","created_at":"2026-06-12T01:09:09.407305+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.15102","citing_title":"Improving Multi-turn Dialogue Consistency with Self-Recall Thinking","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2507.14913","citing_title":"PromptSuite: A Task-Agnostic Framework for Multi-Prompt Generation","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY","json":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY.json","graph_json":"https://pith.science/api/pith-number/TMIDWGUGVKZ6ACPECZTP6JFTDY/graph.json","events_json":"https://pith.science/api/pith-number/TMIDWGUGVKZ6ACPECZTP6JFTDY/events.json","paper":"https://pith.science/paper/TMIDWGUG"},"agent_actions":{"view_html":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY","download_json":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY.json","view_paper":"https://pith.science/paper/TMIDWGUG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22169&json=true","fetch_graph":"https://pith.science/api/pith-number/TMIDWGUGVKZ6ACPECZTP6JFTDY/graph.json","fetch_events":"https://pith.science/api/pith-number/TMIDWGUGVKZ6ACPECZTP6JFTDY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY/action/storage_attestation","attest_author":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY/action/author_attestation","sign_citation":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY/action/citation_signature","submit_replication":"https://pith.science/pith/TMIDWGUGVKZ6ACPECZTP6JFTDY/action/replication_record"}},"created_at":"2026-06-12T01:09:09.407305+00:00","updated_at":"2026-06-12T01:09:09.407305+00:00"}