{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:CH7A2ZXYFSTTRJEVAR537VUMV6","short_pith_number":"pith:CH7A2ZXY","schema_version":"1.0","canonical_sha256":"11fe0d66f82ca738a495047bbfd68caf8cd04ed8074034ae50961be594122872","source":{"kind":"arxiv","id":"2608.04899","version":1},"attestation_state":"computed","paper":{"title":"Evaluation Pitfalls and Sparsity Limitations in LLM-based Confidence Estimates for Classification","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andreas R\\\"uckl\\'e, Elena Merdjanovska, Omar Zaidan","submitted_at":"2026-08-05T14:24:34Z","abstract_excerpt":"Confidence estimation is essential when LLMs are used for classification, indicating when predictions can be trusted. However, common approaches such as verbalization produce extremely sparse outputs. For instance, Qwen3-32B verbalizes only eight unique confidence values on SST-2, with over half being exactly 95%, a pattern we observe consistently across four datasets and two LLMs. Besides limiting practical utility, we show that this sparsity critically affects evaluation: the choice of interpolation in area under the accuracy-rejection curve (AUARC) dramatically alters rankings, with consist"},"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":"2608.04899","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-08-05T14:24:34Z","cross_cats_sorted":[],"title_canon_sha256":"e32165c60c0fcc9f22ac1f67aa41bc5956e649875fcf951ad5c9a8585e34403d","abstract_canon_sha256":"ce88332c3aa473fbf557e690bd9b901abe49f9ede39e17ee48797d33ce2f558a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-06T01:47:39.946643Z","signature_b64":"2xY7iu145e+QY0xaczgmSQu4hPlsgQDAJh0kKFmZ5z/uWozY/ct/LAPbhgeEJ8PazSA+hwfDBnOXcLt0vdd4Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"11fe0d66f82ca738a495047bbfd68caf8cd04ed8074034ae50961be594122872","last_reissued_at":"2026-08-06T01:47:39.945287Z","signature_status":"signed_v1","first_computed_at":"2026-08-06T01:47:39.945287Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Evaluation Pitfalls and Sparsity Limitations in LLM-based Confidence Estimates for Classification","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andreas R\\\"uckl\\'e, Elena Merdjanovska, Omar Zaidan","submitted_at":"2026-08-05T14:24:34Z","abstract_excerpt":"Confidence estimation is essential when LLMs are used for classification, indicating when predictions can be trusted. However, common approaches such as verbalization produce extremely sparse outputs. For instance, Qwen3-32B verbalizes only eight unique confidence values on SST-2, with over half being exactly 95%, a pattern we observe consistently across four datasets and two LLMs. Besides limiting practical utility, we show that this sparsity critically affects evaluation: the choice of interpolation in area under the accuracy-rejection curve (AUARC) dramatically alters rankings, with consist"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.04899","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/2608.04899/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":"2608.04899","created_at":"2026-08-06T01:47:39.947083+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.04899v1","created_at":"2026-08-06T01:47:39.947083+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.04899","created_at":"2026-08-06T01:47:39.947083+00:00"},{"alias_kind":"pith_short_12","alias_value":"CH7A2ZXYFSTT","created_at":"2026-08-06T01:47:39.947083+00:00"},{"alias_kind":"pith_short_16","alias_value":"CH7A2ZXYFSTTRJEV","created_at":"2026-08-06T01:47:39.947083+00:00"},{"alias_kind":"pith_short_8","alias_value":"CH7A2ZXY","created_at":"2026-08-06T01:47:39.947083+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/CH7A2ZXYFSTTRJEVAR537VUMV6","json":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6.json","graph_json":"https://pith.science/api/pith-number/CH7A2ZXYFSTTRJEVAR537VUMV6/graph.json","events_json":"https://pith.science/api/pith-number/CH7A2ZXYFSTTRJEVAR537VUMV6/events.json","paper":"https://pith.science/paper/CH7A2ZXY"},"agent_actions":{"view_html":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6","download_json":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6.json","view_paper":"https://pith.science/paper/CH7A2ZXY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.04899&json=true","fetch_graph":"https://pith.science/api/pith-number/CH7A2ZXYFSTTRJEVAR537VUMV6/graph.json","fetch_events":"https://pith.science/api/pith-number/CH7A2ZXYFSTTRJEVAR537VUMV6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6/action/storage_attestation","attest_author":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6/action/author_attestation","sign_citation":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6/action/citation_signature","submit_replication":"https://pith.science/pith/CH7A2ZXYFSTTRJEVAR537VUMV6/action/replication_record"}},"created_at":"2026-08-06T01:47:39.947083+00:00","updated_at":"2026-08-06T01:47:39.947083+00:00"}