{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:B4WNMLNOCAKRWVHUUPNHDAFSRF","short_pith_number":"pith:B4WNMLNO","schema_version":"1.0","canonical_sha256":"0f2cd62dae10151b54f4a3da7180b2897fa141d481126cd3ce35291c07a25a37","source":{"kind":"arxiv","id":"2605.29889","version":1},"attestation_state":"computed","paper":{"title":"Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Berardino Como, David Fraile Navarro, Jialei Sheng, Shlomo Berkovsky, Soundariya Ananthan","submitted_at":"2026-05-28T13:14:17Z","abstract_excerpt":"Patient-voiced clinical-triage benchmarks report high under-triage rates for consumer LLMs for constrained multiple-choice output, yet the same cases score differently with free-text. We ask whether output format changes the model's \\emph{clinical representation} or only the mapping from a preserved representation to an answer. Using sparse-autoencoder (SAE) features in Gemma 3 4B/12B IT and Qwen3-8B, we find the same medical features fire on the shared clinical narrative under both formats but go {silent} at the multiple-choice decision token in all the cases at every model. Three independent"},"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":"2605.29889","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-28T13:14:17Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b1e9b2a3d5ad9a382842e46ed2114dcd21cf5d1f4ea1b4d11145f9d3bf940d59","abstract_canon_sha256":"aa6c9e5836f8a42cfe89bfdb36422c0a4808fd7d7b91454b978814ec10b973ec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:05:57.955037Z","signature_b64":"G7o9SxXsQUJjClfQb0SP9/rQQNQbEFFUoRxi0IPm+/542kbs/WD2crtwRuWz4wJG/H+KeZiakJ2lVqcsD1b2CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f2cd62dae10151b54f4a3da7180b2897fa141d481126cd3ce35291c07a25a37","last_reissued_at":"2026-05-29T02:05:57.954500Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:05:57.954500Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Berardino Como, David Fraile Navarro, Jialei Sheng, Shlomo Berkovsky, Soundariya Ananthan","submitted_at":"2026-05-28T13:14:17Z","abstract_excerpt":"Patient-voiced clinical-triage benchmarks report high under-triage rates for consumer LLMs for constrained multiple-choice output, yet the same cases score differently with free-text. We ask whether output format changes the model's \\emph{clinical representation} or only the mapping from a preserved representation to an answer. Using sparse-autoencoder (SAE) features in Gemma 3 4B/12B IT and Qwen3-8B, we find the same medical features fire on the shared clinical narrative under both formats but go {silent} at the multiple-choice decision token in all the cases at every model. Three independent"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.29889","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/2605.29889/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":"2605.29889","created_at":"2026-05-29T02:05:57.954578+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.29889v1","created_at":"2026-05-29T02:05:57.954578+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.29889","created_at":"2026-05-29T02:05:57.954578+00:00"},{"alias_kind":"pith_short_12","alias_value":"B4WNMLNOCAKR","created_at":"2026-05-29T02:05:57.954578+00:00"},{"alias_kind":"pith_short_16","alias_value":"B4WNMLNOCAKRWVHU","created_at":"2026-05-29T02:05:57.954578+00:00"},{"alias_kind":"pith_short_8","alias_value":"B4WNMLNO","created_at":"2026-05-29T02:05:57.954578+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/B4WNMLNOCAKRWVHUUPNHDAFSRF","json":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF.json","graph_json":"https://pith.science/api/pith-number/B4WNMLNOCAKRWVHUUPNHDAFSRF/graph.json","events_json":"https://pith.science/api/pith-number/B4WNMLNOCAKRWVHUUPNHDAFSRF/events.json","paper":"https://pith.science/paper/B4WNMLNO"},"agent_actions":{"view_html":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF","download_json":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF.json","view_paper":"https://pith.science/paper/B4WNMLNO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.29889&json=true","fetch_graph":"https://pith.science/api/pith-number/B4WNMLNOCAKRWVHUUPNHDAFSRF/graph.json","fetch_events":"https://pith.science/api/pith-number/B4WNMLNOCAKRWVHUUPNHDAFSRF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF/action/storage_attestation","attest_author":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF/action/author_attestation","sign_citation":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF/action/citation_signature","submit_replication":"https://pith.science/pith/B4WNMLNOCAKRWVHUUPNHDAFSRF/action/replication_record"}},"created_at":"2026-05-29T02:05:57.954578+00:00","updated_at":"2026-05-29T02:05:57.954578+00:00"}