{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GYD5P3Q5XBDDSHNTZR5CXGTVO6","short_pith_number":"pith:GYD5P3Q5","schema_version":"1.0","canonical_sha256":"3607d7ee1db846391db3cc7a2b9a7577b46e0d44e8ddf3a81ad8ff3479804e13","source":{"kind":"arxiv","id":"2607.18006","version":1},"attestation_state":"computed","paper":{"title":"MADA-RL: Multi-Agent Debate-Aware Reinforcement Learning for Parameter-Efficient Reasoning in Compact Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.MA"],"primary_cat":"cs.LG","authors_text":"Cuong Xuan Chu, Evgeny Kharlamov, Martino M. L. Pulici, Volker Tresp, Yunpu Ma, Zifeng Ding","submitted_at":"2026-07-20T14:38:00Z","abstract_excerpt":"Large language models achieve strong reasoning performance, but often at prohibitive training cost - a challenge that is especially acute for compact models ($\\leq 4 \\, \\mathrm{B}$ parameters) trained under limited budgets. We introduce MADA-RL, a post-training framework that specializes compact models into generator and critic roles and trains them with a debate-aware learning signal, fine-tuning only a small subset of parameters via LoRA adapters. Our central contribution is a counterfactual critic advantage: a dynamic, role-conditioned baseline that redefines the critic's advantage as its r"},"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":"2607.18006","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-20T14:38:00Z","cross_cats_sorted":["cs.AI","cs.CL","cs.MA"],"title_canon_sha256":"55e2cf8075eef22e724150159360e67b2b5d794f569d2806f20e34610efeb276","abstract_canon_sha256":"372207c3b9e110cd1aa0db915fe204242db964cecf5a65459244f9dc0137d227"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-21T02:22:11.467239Z","signature_b64":"iHrKYH6O5mRQNAy5XpCgSKOixSOGljHjQ96DSFwTIunQsvn5XnON3F9p11hqO30Qp4jBkSlWUE9GJAT+eEVwDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3607d7ee1db846391db3cc7a2b9a7577b46e0d44e8ddf3a81ad8ff3479804e13","last_reissued_at":"2026-07-21T02:22:11.466395Z","signature_status":"signed_v1","first_computed_at":"2026-07-21T02:22:11.466395Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MADA-RL: Multi-Agent Debate-Aware Reinforcement Learning for Parameter-Efficient Reasoning in Compact Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.MA"],"primary_cat":"cs.LG","authors_text":"Cuong Xuan Chu, Evgeny Kharlamov, Martino M. L. Pulici, Volker Tresp, Yunpu Ma, Zifeng Ding","submitted_at":"2026-07-20T14:38:00Z","abstract_excerpt":"Large language models achieve strong reasoning performance, but often at prohibitive training cost - a challenge that is especially acute for compact models ($\\leq 4 \\, \\mathrm{B}$ parameters) trained under limited budgets. We introduce MADA-RL, a post-training framework that specializes compact models into generator and critic roles and trains them with a debate-aware learning signal, fine-tuning only a small subset of parameters via LoRA adapters. Our central contribution is a counterfactual critic advantage: a dynamic, role-conditioned baseline that redefines the critic's advantage as its r"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.18006","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/2607.18006/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":"2607.18006","created_at":"2026-07-21T02:22:11.466836+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.18006v1","created_at":"2026-07-21T02:22:11.466836+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.18006","created_at":"2026-07-21T02:22:11.466836+00:00"},{"alias_kind":"pith_short_12","alias_value":"GYD5P3Q5XBDD","created_at":"2026-07-21T02:22:11.466836+00:00"},{"alias_kind":"pith_short_16","alias_value":"GYD5P3Q5XBDDSHNT","created_at":"2026-07-21T02:22:11.466836+00:00"},{"alias_kind":"pith_short_8","alias_value":"GYD5P3Q5","created_at":"2026-07-21T02:22:11.466836+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/GYD5P3Q5XBDDSHNTZR5CXGTVO6","json":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6.json","graph_json":"https://pith.science/api/pith-number/GYD5P3Q5XBDDSHNTZR5CXGTVO6/graph.json","events_json":"https://pith.science/api/pith-number/GYD5P3Q5XBDDSHNTZR5CXGTVO6/events.json","paper":"https://pith.science/paper/GYD5P3Q5"},"agent_actions":{"view_html":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6","download_json":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6.json","view_paper":"https://pith.science/paper/GYD5P3Q5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.18006&json=true","fetch_graph":"https://pith.science/api/pith-number/GYD5P3Q5XBDDSHNTZR5CXGTVO6/graph.json","fetch_events":"https://pith.science/api/pith-number/GYD5P3Q5XBDDSHNTZR5CXGTVO6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6/action/storage_attestation","attest_author":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6/action/author_attestation","sign_citation":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6/action/citation_signature","submit_replication":"https://pith.science/pith/GYD5P3Q5XBDDSHNTZR5CXGTVO6/action/replication_record"}},"created_at":"2026-07-21T02:22:11.466836+00:00","updated_at":"2026-07-21T02:22:11.466836+00:00"}