{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OFT3HZQ7BVZKEDQE5LVMKDHIOY","short_pith_number":"pith:OFT3HZQ7","schema_version":"1.0","canonical_sha256":"7167b3e61f0d72a20e04eaeac50ce876153f4c348d46c5c4fa60197fca09efed","source":{"kind":"arxiv","id":"2606.06058","version":1},"attestation_state":"computed","paper":{"title":"MDP-GRPO: Stabilized Group Relative Policy Optimization for Multi-Constraint Instruction Following","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Heshaam Faili, Mohammad Javad Dousti, Mohammad Mahdi Salmani-Zarchi, Zahra Rahimi","submitted_at":"2026-06-04T11:58:59Z","abstract_excerpt":"Reinforcement learning with verifiable rewards is ideal for multi-constraint instruction following, yet standard group-relative policy optimization (GRPO) becomes unstable under discrete, low-dispersion rewards, where within-group reward distributions are frequently homogeneous. We identify and formalize three pathologies of z-score group normalization in this regime: low-variance amplification, mean-centering blindness, and zero-variance collapse. To address them, we propose MDP-GRPO, which stabilizes learning through (1) multi-temperature sampling to increase reward dispersion, (2) dual-anch"},"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":"2606.06058","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-04T11:58:59Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"385b174d97b67f97953c411c616d24772f92d5ff5ae8d68b8bfdb6c065de6fd0","abstract_canon_sha256":"30b141d9574ac9f637ed0bc5d879aa9a5fe7fe9930444e7b3326427693830873"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:31.761205Z","signature_b64":"c8DWSaYTcH06RwpyrolLqG6YTdYifBx7z0bYPenlxrFGCIUgZ1jsRNy2YQsNSmAJhRUzGuhOaVxnVhta1yqPCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7167b3e61f0d72a20e04eaeac50ce876153f4c348d46c5c4fa60197fca09efed","last_reissued_at":"2026-06-05T01:15:31.760806Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:31.760806Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MDP-GRPO: Stabilized Group Relative Policy Optimization for Multi-Constraint Instruction Following","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Heshaam Faili, Mohammad Javad Dousti, Mohammad Mahdi Salmani-Zarchi, Zahra Rahimi","submitted_at":"2026-06-04T11:58:59Z","abstract_excerpt":"Reinforcement learning with verifiable rewards is ideal for multi-constraint instruction following, yet standard group-relative policy optimization (GRPO) becomes unstable under discrete, low-dispersion rewards, where within-group reward distributions are frequently homogeneous. We identify and formalize three pathologies of z-score group normalization in this regime: low-variance amplification, mean-centering blindness, and zero-variance collapse. To address them, we propose MDP-GRPO, which stabilizes learning through (1) multi-temperature sampling to increase reward dispersion, (2) dual-anch"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.06058","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/2606.06058/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":"2606.06058","created_at":"2026-06-05T01:15:31.760868+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.06058v1","created_at":"2026-06-05T01:15:31.760868+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.06058","created_at":"2026-06-05T01:15:31.760868+00:00"},{"alias_kind":"pith_short_12","alias_value":"OFT3HZQ7BVZK","created_at":"2026-06-05T01:15:31.760868+00:00"},{"alias_kind":"pith_short_16","alias_value":"OFT3HZQ7BVZKEDQE","created_at":"2026-06-05T01:15:31.760868+00:00"},{"alias_kind":"pith_short_8","alias_value":"OFT3HZQ7","created_at":"2026-06-05T01:15:31.760868+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/OFT3HZQ7BVZKEDQE5LVMKDHIOY","json":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY.json","graph_json":"https://pith.science/api/pith-number/OFT3HZQ7BVZKEDQE5LVMKDHIOY/graph.json","events_json":"https://pith.science/api/pith-number/OFT3HZQ7BVZKEDQE5LVMKDHIOY/events.json","paper":"https://pith.science/paper/OFT3HZQ7"},"agent_actions":{"view_html":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY","download_json":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY.json","view_paper":"https://pith.science/paper/OFT3HZQ7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.06058&json=true","fetch_graph":"https://pith.science/api/pith-number/OFT3HZQ7BVZKEDQE5LVMKDHIOY/graph.json","fetch_events":"https://pith.science/api/pith-number/OFT3HZQ7BVZKEDQE5LVMKDHIOY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY/action/storage_attestation","attest_author":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY/action/author_attestation","sign_citation":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY/action/citation_signature","submit_replication":"https://pith.science/pith/OFT3HZQ7BVZKEDQE5LVMKDHIOY/action/replication_record"}},"created_at":"2026-06-05T01:15:31.760868+00:00","updated_at":"2026-06-05T01:15:31.760868+00:00"}