{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MLOUKLCODKZ4NCRO5M4NYKOLOF","short_pith_number":"pith:MLOUKLCO","schema_version":"1.0","canonical_sha256":"62dd452c4e1ab3c68a2eeb38dc29cb714825b2dfbbe1c3db235971db7308284a","source":{"kind":"arxiv","id":"2606.14530","version":2},"attestation_state":"computed","paper":{"title":"Code Correctness Signals in LLM Hidden States: Pre-Generation Probing and Repair Geometry","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Carlo Di Cicco","submitted_at":"2026-06-12T15:06:15Z","abstract_excerpt":"Large language models encode rich information in their hidden states. This work asks whether code correctness is legible in the hidden states of Qwen3-4B-Instruct-2507, before it generates and as it repairs a failed attempt, studied on 444 LiveCodeBench tasks. It reports two findings connected by a single confound-control tool: residualization. First, the correctness of the model's first-attempt code is linearly decodable from the prompt-final hidden state, with a leakage-free held-out AUC of 0.955 +/- 0.006 across 50 outer splits. After the linear effect of prompt length is removed from each "},"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":"2606.14530","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-12T15:06:15Z","cross_cats_sorted":[],"title_canon_sha256":"52d2336dd681244be07cd4c064626de0c124b86596db1cf0c6522f6a14b63f75","abstract_canon_sha256":"882ab3e33cc7ea821122b4aa1dae202940104eac1fe132d3e7db047db2064470"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-13T01:18:14.846525Z","signature_b64":"TtJH3PmaAsJLfiJJj41/MNP3MaMFwXH/hRSNFPFwrLx84GrTlFlIlJ8/bcjKyBRsHD+BAME16JlP5tiirFi4Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62dd452c4e1ab3c68a2eeb38dc29cb714825b2dfbbe1c3db235971db7308284a","last_reissued_at":"2026-07-13T01:18:14.844960Z","signature_status":"signed_v1","first_computed_at":"2026-07-13T01:18:14.844960Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Code Correctness Signals in LLM Hidden States: Pre-Generation Probing and Repair Geometry","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Carlo Di Cicco","submitted_at":"2026-06-12T15:06:15Z","abstract_excerpt":"Large language models encode rich information in their hidden states. This work asks whether code correctness is legible in the hidden states of Qwen3-4B-Instruct-2507, before it generates and as it repairs a failed attempt, studied on 444 LiveCodeBench tasks. It reports two findings connected by a single confound-control tool: residualization. First, the correctness of the model's first-attempt code is linearly decodable from the prompt-final hidden state, with a leakage-free held-out AUC of 0.955 +/- 0.006 across 50 outer splits. After the linear effect of prompt length is removed from each "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.14530","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/2606.14530/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":"2606.14530","created_at":"2026-07-13T01:18:14.845695+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.14530v2","created_at":"2026-07-13T01:18:14.845695+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.14530","created_at":"2026-07-13T01:18:14.845695+00:00"},{"alias_kind":"pith_short_12","alias_value":"MLOUKLCODKZ4","created_at":"2026-07-13T01:18:14.845695+00:00"},{"alias_kind":"pith_short_16","alias_value":"MLOUKLCODKZ4NCRO","created_at":"2026-07-13T01:18:14.845695+00:00"},{"alias_kind":"pith_short_8","alias_value":"MLOUKLCO","created_at":"2026-07-13T01:18:14.845695+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05188","citing_title":"Latent Programming Horizons in Coding Agents","ref_index":10,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF","json":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF.json","graph_json":"https://pith.science/api/pith-number/MLOUKLCODKZ4NCRO5M4NYKOLOF/graph.json","events_json":"https://pith.science/api/pith-number/MLOUKLCODKZ4NCRO5M4NYKOLOF/events.json","paper":"https://pith.science/paper/MLOUKLCO"},"agent_actions":{"view_html":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF","download_json":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF.json","view_paper":"https://pith.science/paper/MLOUKLCO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.14530&json=true","fetch_graph":"https://pith.science/api/pith-number/MLOUKLCODKZ4NCRO5M4NYKOLOF/graph.json","fetch_events":"https://pith.science/api/pith-number/MLOUKLCODKZ4NCRO5M4NYKOLOF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF/action/storage_attestation","attest_author":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF/action/author_attestation","sign_citation":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF/action/citation_signature","submit_replication":"https://pith.science/pith/MLOUKLCODKZ4NCRO5M4NYKOLOF/action/replication_record"}},"created_at":"2026-07-13T01:18:14.845695+00:00","updated_at":"2026-07-13T01:18:14.845695+00:00"}