{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZJANNJBFD5DNWTOM4HGYULHEFD","short_pith_number":"pith:ZJANNJBF","schema_version":"1.0","canonical_sha256":"ca40d6a4251f46db4dcce1cd8a2ce428d0b8fe2f23dfb7fe9b70f6fb8603c9c4","source":{"kind":"arxiv","id":"2607.16244","version":1},"attestation_state":"computed","paper":{"title":"CIGPO: Contextual Information-Gain Policy Optimization for Multi-Turn Evidence-Reading LLM Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hao Dou","submitted_at":"2026-06-26T11:50:29Z","abstract_excerpt":"Training multi-turn evidence-reading agents with outcome-only reinforcement learning is unstable because intermediate turns receive little direct credit. In HotpotQA experiments with Qwen2.5-3B-Instruct, GRPO initially improves (standard F1 0.430) but subsequently collapses to 100% format-violating outputs. Training-log diagnosis reveals a zero-advantage lock-in mechanism: all sampled trajectories receive the minimum format penalty (-2.0), group-relative advantages vanish, and the policy-gradient loss becomes zero--an optimization deadlock. We propose a variance-injection strategy: by assignin"},"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":"2607.16244","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-26T11:50:29Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"128806b972b05c45b2c4f051329c2341d319e4a2c4686fa3074805856d50b9f1","abstract_canon_sha256":"a9af0d70757cee29ce9f7cc59785ebd2c06a62b5e1c749bb743584cfcb72aea2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-21T00:20:07.846259Z","signature_b64":"cCLw/8PueICiTw9B9xFNaq4+3gMaC8jYdnY6xVmN0t1ow8Jwsfx+v+v7VsFPo4M3piFCi+PcFpJ4t5ufHsFYCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ca40d6a4251f46db4dcce1cd8a2ce428d0b8fe2f23dfb7fe9b70f6fb8603c9c4","last_reissued_at":"2026-07-21T00:20:07.845381Z","signature_status":"signed_v1","first_computed_at":"2026-07-21T00:20:07.845381Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CIGPO: Contextual Information-Gain Policy Optimization for Multi-Turn Evidence-Reading LLM Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hao Dou","submitted_at":"2026-06-26T11:50:29Z","abstract_excerpt":"Training multi-turn evidence-reading agents with outcome-only reinforcement learning is unstable because intermediate turns receive little direct credit. In HotpotQA experiments with Qwen2.5-3B-Instruct, GRPO initially improves (standard F1 0.430) but subsequently collapses to 100% format-violating outputs. Training-log diagnosis reveals a zero-advantage lock-in mechanism: all sampled trajectories receive the minimum format penalty (-2.0), group-relative advantages vanish, and the policy-gradient loss becomes zero--an optimization deadlock. We propose a variance-injection strategy: by assignin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.16244","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/2607.16244/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":"2607.16244","created_at":"2026-07-21T00:20:07.845847+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.16244v1","created_at":"2026-07-21T00:20:07.845847+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.16244","created_at":"2026-07-21T00:20:07.845847+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZJANNJBFD5DN","created_at":"2026-07-21T00:20:07.845847+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZJANNJBFD5DNWTOM","created_at":"2026-07-21T00:20:07.845847+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZJANNJBF","created_at":"2026-07-21T00:20:07.845847+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/ZJANNJBFD5DNWTOM4HGYULHEFD","json":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD.json","graph_json":"https://pith.science/api/pith-number/ZJANNJBFD5DNWTOM4HGYULHEFD/graph.json","events_json":"https://pith.science/api/pith-number/ZJANNJBFD5DNWTOM4HGYULHEFD/events.json","paper":"https://pith.science/paper/ZJANNJBF"},"agent_actions":{"view_html":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD","download_json":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD.json","view_paper":"https://pith.science/paper/ZJANNJBF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.16244&json=true","fetch_graph":"https://pith.science/api/pith-number/ZJANNJBFD5DNWTOM4HGYULHEFD/graph.json","fetch_events":"https://pith.science/api/pith-number/ZJANNJBFD5DNWTOM4HGYULHEFD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD/action/storage_attestation","attest_author":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD/action/author_attestation","sign_citation":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD/action/citation_signature","submit_replication":"https://pith.science/pith/ZJANNJBFD5DNWTOM4HGYULHEFD/action/replication_record"}},"created_at":"2026-07-21T00:20:07.845847+00:00","updated_at":"2026-07-21T00:20:07.845847+00:00"}