{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Q24F4WTDUUZJBKFTRIHYLH3REB","short_pith_number":"pith:Q24F4WTD","schema_version":"1.0","canonical_sha256":"86b85e5a63a53290a8b38a0f859f71207df16b4f2642e6762684fdad869f3fba","source":{"kind":"arxiv","id":"2605.22544","version":1},"attestation_state":"computed","paper":{"title":"One prompt is not enough: Instruction Sensitivity Undermines Embedding Model Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IR"],"primary_cat":"cs.CL","authors_text":"Kenneth Enevoldsen, Yevhen Kostiuk","submitted_at":"2026-05-21T14:27:46Z","abstract_excerpt":"Instruction embedding models have become common among state-of-the-art models, however are evaluated using a single prompt per task. The single-point evaluation ignores a main problem of the instruction-based approach namely: sensitivity to the phrasing of the instruction. We present an empirical study of prompt sensitivity across 6 embedding models, 11 datasets, and 15 task-specific prompts per dataset, a total of 990. We show that reported scores misrepresent the distribution of scores over plausible prompts. The default prompt can both systematically understate or overstate performance. Fur"},"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":"2605.22544","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-21T14:27:46Z","cross_cats_sorted":["cs.IR"],"title_canon_sha256":"576af77fe98996cd3d6760a49ddcff94fdd18a1544ad9ff8b8792152ee644a68","abstract_canon_sha256":"4e20f71ee5616c5cc507607d492961abe5750987a67f1c05c55ce506af380193"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:56.576770Z","signature_b64":"lb3b1tlK+gttWOE2wfiMBRANUFJtXfpBl1iNok68QmS5oCa1geaxRuHdmnUK57BTKfzz9uvPZkWUUe+Kd6W7BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"86b85e5a63a53290a8b38a0f859f71207df16b4f2642e6762684fdad869f3fba","last_reissued_at":"2026-05-22T01:04:56.575939Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:56.575939Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"One prompt is not enough: Instruction Sensitivity Undermines Embedding Model Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IR"],"primary_cat":"cs.CL","authors_text":"Kenneth Enevoldsen, Yevhen Kostiuk","submitted_at":"2026-05-21T14:27:46Z","abstract_excerpt":"Instruction embedding models have become common among state-of-the-art models, however are evaluated using a single prompt per task. The single-point evaluation ignores a main problem of the instruction-based approach namely: sensitivity to the phrasing of the instruction. We present an empirical study of prompt sensitivity across 6 embedding models, 11 datasets, and 15 task-specific prompts per dataset, a total of 990. We show that reported scores misrepresent the distribution of scores over plausible prompts. The default prompt can both systematically understate or overstate performance. Fur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22544","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/2605.22544/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":"2605.22544","created_at":"2026-05-22T01:04:56.576069+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22544v1","created_at":"2026-05-22T01:04:56.576069+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22544","created_at":"2026-05-22T01:04:56.576069+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q24F4WTDUUZJ","created_at":"2026-05-22T01:04:56.576069+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q24F4WTDUUZJBKFT","created_at":"2026-05-22T01:04:56.576069+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q24F4WTD","created_at":"2026-05-22T01:04:56.576069+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/Q24F4WTDUUZJBKFTRIHYLH3REB","json":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB.json","graph_json":"https://pith.science/api/pith-number/Q24F4WTDUUZJBKFTRIHYLH3REB/graph.json","events_json":"https://pith.science/api/pith-number/Q24F4WTDUUZJBKFTRIHYLH3REB/events.json","paper":"https://pith.science/paper/Q24F4WTD"},"agent_actions":{"view_html":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB","download_json":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB.json","view_paper":"https://pith.science/paper/Q24F4WTD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22544&json=true","fetch_graph":"https://pith.science/api/pith-number/Q24F4WTDUUZJBKFTRIHYLH3REB/graph.json","fetch_events":"https://pith.science/api/pith-number/Q24F4WTDUUZJBKFTRIHYLH3REB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB/action/storage_attestation","attest_author":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB/action/author_attestation","sign_citation":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB/action/citation_signature","submit_replication":"https://pith.science/pith/Q24F4WTDUUZJBKFTRIHYLH3REB/action/replication_record"}},"created_at":"2026-05-22T01:04:56.576069+00:00","updated_at":"2026-05-22T01:04:56.576069+00:00"}