{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:B232T5UT6PPUFSQAU2JAKTAO6N","short_pith_number":"pith:B232T5UT","canonical_record":{"source":{"id":"2603.09117","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-10T02:47:59Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"c37876e30466eadc3601baea3bd899d61a4245cc748545d8b2e24b06d8c3cc91","abstract_canon_sha256":"8c44414232b4e60e063d3c0d487e7cb4665ee0263df0305a43a74d82b447ca24"},"schema_version":"1.0"},"canonical_sha256":"0eb7a9f693f3df42ca00a692054c0ef36363aa2efa9de35dcea50a62f4ad2e40","source":{"kind":"arxiv","id":"2603.09117","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.09117","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"arxiv_version","alias_value":"2603.09117v3","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.09117","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_12","alias_value":"B232T5UT6PPU","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_16","alias_value":"B232T5UT6PPUFSQA","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_8","alias_value":"B232T5UT","created_at":"2026-05-28T01:04:38Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:B232T5UT6PPUFSQAU2JAKTAO6N","target":"record","payload":{"canonical_record":{"source":{"id":"2603.09117","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-10T02:47:59Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"c37876e30466eadc3601baea3bd899d61a4245cc748545d8b2e24b06d8c3cc91","abstract_canon_sha256":"8c44414232b4e60e063d3c0d487e7cb4665ee0263df0305a43a74d82b447ca24"},"schema_version":"1.0"},"canonical_sha256":"0eb7a9f693f3df42ca00a692054c0ef36363aa2efa9de35dcea50a62f4ad2e40","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:38.117632Z","signature_b64":"L2rGqh8OOckbB9wmlGFhzHIdvgfbBc8DL6869wAhX008umWOy68WxDzyRtYXCRDYFkEUydhmsayFbTZq6TH4CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0eb7a9f693f3df42ca00a692054c0ef36363aa2efa9de35dcea50a62f4ad2e40","last_reissued_at":"2026-05-28T01:04:38.117064Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:38.117064Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2603.09117","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-28T01:04:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"LSAQMgvsC6IBj0N8C/kFut+FrAziYf+PXMkm5aooFSe+LSgttxDEfOgwqAzs7wgN186T0e5UMQzNTcrYHt5CDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T12:26:33.752422Z"},"content_sha256":"c498f93d5503025c6505e34b78784f6435250187d1c8f368842eae66bc9fcce3","schema_version":"1.0","event_id":"sha256:c498f93d5503025c6505e34b78784f6435250187d1c8f368842eae66bc9fcce3"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:B232T5UT6PPUFSQAU2JAKTAO6N","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Boxi Cao, Hongyu Lin, Jinglin Yang, Le Sun, Min He, Xianpei Han, Xueru Wen, Yaojie Lu, Zhengzhao Ma","submitted_at":"2026-03-10T02:47:59Z","abstract_excerpt":"Reinforcement Learning from Verifiable Rewards (RLVR) significantly enhances large language models (LLMs) reasoning but severely suffers from calibration degeneration, where models become excessively over-confident in incorrect answers. Previous studies devote to directly incorporating calibration objective into existing optimization target. However, our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. Building on this insight, we propose DCPO, a simple yet effective fra"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the proposed decoupling in DCPO can be implemented without introducing new optimization instabilities or unintended side effects on other model behaviors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DCPO decouples reasoning optimization from calibration in RLVR to fix overconfidence in LLMs without losing accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2767baea4696a13d11db56885f6cb8d75913cebff050c4f0b725b38001e9fe4e"},"source":{"id":"2603.09117","kind":"arxiv","version":3},"verdict":{"id":"330c10c0-88cc-4ef5-b559-2a5b73e6afa6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T13:45:40.413602Z","strongest_claim":"Our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue.","one_line_summary":"DCPO decouples reasoning optimization from calibration in RLVR to fix overconfidence in LLMs without losing accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the proposed decoupling in DCPO can be implemented without introducing new optimization instabilities or unintended side effects on other model behaviors.","pith_extraction_headline":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.09117/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"},"verdict_id":"330c10c0-88cc-4ef5-b559-2a5b73e6afa6"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-28T01:04:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"p1OHqHysrJquvTx9sdOF1ka0BwgNbv5fO8zr1VgWPtfRcFolr+CtfAWMhG9sbdwLLom0rUHNWqdbP+lceKK8CQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T12:26:33.753476Z"},"content_sha256":"671777f28b3f6596b3c55fb4ee17617dac26ef8bd36f353910f2af4e5fe3a54c","schema_version":"1.0","event_id":"sha256:671777f28b3f6596b3c55fb4ee17617dac26ef8bd36f353910f2af4e5fe3a54c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/B232T5UT6PPUFSQAU2JAKTAO6N/bundle.json","state_url":"https://pith.science/pith/B232T5UT6PPUFSQAU2JAKTAO6N/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/B232T5UT6PPUFSQAU2JAKTAO6N/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-10T12:26:33Z","links":{"resolver":"https://pith.science/pith/B232T5UT6PPUFSQAU2JAKTAO6N","bundle":"https://pith.science/pith/B232T5UT6PPUFSQAU2JAKTAO6N/bundle.json","state":"https://pith.science/pith/B232T5UT6PPUFSQAU2JAKTAO6N/state.json","well_known_bundle":"https://pith.science/.well-known/pith/B232T5UT6PPUFSQAU2JAKTAO6N/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:B232T5UT6PPUFSQAU2JAKTAO6N","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"8c44414232b4e60e063d3c0d487e7cb4665ee0263df0305a43a74d82b447ca24","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-10T02:47:59Z","title_canon_sha256":"c37876e30466eadc3601baea3bd899d61a4245cc748545d8b2e24b06d8c3cc91"},"schema_version":"1.0","source":{"id":"2603.09117","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.09117","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"arxiv_version","alias_value":"2603.09117v3","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.09117","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_12","alias_value":"B232T5UT6PPU","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_16","alias_value":"B232T5UT6PPUFSQA","created_at":"2026-05-28T01:04:38Z"},{"alias_kind":"pith_short_8","alias_value":"B232T5UT","created_at":"2026-05-28T01:04:38Z"}],"graph_snapshots":[{"event_id":"sha256:671777f28b3f6596b3c55fb4ee17617dac26ef8bd36f353910f2af4e5fe3a54c","target":"graph","created_at":"2026-05-28T01:04:38Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the proposed decoupling in DCPO can be implemented without introducing new optimization instabilities or unintended side effects on other model behaviors."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"DCPO decouples reasoning optimization from calibration in RLVR to fix overconfidence in LLMs without losing accuracy."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO."}],"snapshot_sha256":"2767baea4696a13d11db56885f6cb8d75913cebff050c4f0b725b38001e9fe4e"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.09117/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement Learning from Verifiable Rewards (RLVR) significantly enhances large language models (LLMs) reasoning but severely suffers from calibration degeneration, where models become excessively over-confident in incorrect answers. Previous studies devote to directly incorporating calibration objective into existing optimization target. However, our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. Building on this insight, we propose DCPO, a simple yet effective fra","authors_text":"Boxi Cao, Hongyu Lin, Jinglin Yang, Le Sun, Min He, Xianpei Han, Xueru Wen, Yaojie Lu, Zhengzhao Ma","cross_cats":["cs.AI","cs.CL"],"headline":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-10T02:47:59Z","title":"Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.09117","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T13:45:40.413602Z","id":"330c10c0-88cc-4ef5-b559-2a5b73e6afa6","model_set":{"reader":"grok-4.3"},"one_line_summary":"DCPO decouples reasoning optimization from calibration in RLVR to fix overconfidence in LLMs without losing accuracy.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A gradient conflict between accuracy and calibration in RLVR is resolved by decoupling the objectives in DCPO.","strongest_claim":"Our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue.","weakest_assumption":"That the proposed decoupling in DCPO can be implemented without introducing new optimization instabilities or unintended side effects on other model behaviors."}},"verdict_id":"330c10c0-88cc-4ef5-b559-2a5b73e6afa6"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c498f93d5503025c6505e34b78784f6435250187d1c8f368842eae66bc9fcce3","target":"record","created_at":"2026-05-28T01:04:38Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"8c44414232b4e60e063d3c0d487e7cb4665ee0263df0305a43a74d82b447ca24","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-10T02:47:59Z","title_canon_sha256":"c37876e30466eadc3601baea3bd899d61a4245cc748545d8b2e24b06d8c3cc91"},"schema_version":"1.0","source":{"id":"2603.09117","kind":"arxiv","version":3}},"canonical_sha256":"0eb7a9f693f3df42ca00a692054c0ef36363aa2efa9de35dcea50a62f4ad2e40","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0eb7a9f693f3df42ca00a692054c0ef36363aa2efa9de35dcea50a62f4ad2e40","first_computed_at":"2026-05-28T01:04:38.117064Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-28T01:04:38.117064Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"L2rGqh8OOckbB9wmlGFhzHIdvgfbBc8DL6869wAhX008umWOy68WxDzyRtYXCRDYFkEUydhmsayFbTZq6TH4CA==","signature_status":"signed_v1","signed_at":"2026-05-28T01:04:38.117632Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.09117","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c498f93d5503025c6505e34b78784f6435250187d1c8f368842eae66bc9fcce3","sha256:671777f28b3f6596b3c55fb4ee17617dac26ef8bd36f353910f2af4e5fe3a54c"],"state_sha256":"227f24cca4e1d1984abab78b70041b57e51d2ad72eebc784f128dd68717e6313"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"tzuK4dhypm2JzQw64X/OpD9+9KZM4+zTNqwAQrboqAEwV3O9ofKQx2ThzisotE3d6GqBPOLY6XD6lHr6e4mhDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-10T12:26:33.758468Z","bundle_sha256":"0b8256e0ef08246c765f4a3e0dd6368d7780070c6508868ca40aec347cccfabf"}}