{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZOO2NWSMVCHHTJG7K6LAY66X4A","short_pith_number":"pith:ZOO2NWSM","schema_version":"1.0","canonical_sha256":"cb9da6da4ca88e79a4df57960c7bd7e03848d275b8d375c633479abb351c0358","source":{"kind":"arxiv","id":"2607.01490","version":1},"attestation_state":"computed","paper":{"title":"Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Francis Bach, Gabriel Synnaeve, Juliette Decugis, Sean O'Brien, Taco Cohen","submitted_at":"2026-07-01T21:39:19Z","abstract_excerpt":"Reinforcement learning post-training dramatically improves LLM reasoning, but suffers from training instability and diversity collapse. Advantage functions offer an appealing fix: they reshape the training objective, reweight which rollouts drive learning, and are trivial to implement. Yet a proliferation of methods makes it unclear which advantage to use and when. We cut through the confusion with a unifying framework that decomposes any advantage into its positive and negative gradient mass along two orthogonal axes. On the sign axis, imbalanced updates collapse either entropy or weight geom"},"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":"2607.01490","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-01T21:39:19Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0536d366e263af226d343910c4732da138b75b291886b2f14fba3953cd5b4396","abstract_canon_sha256":"f15ae361b004cfd956d09abddfc5b4875924fa3308863fa6850512199a63154c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-03T00:17:02.222357Z","signature_b64":"oYTv5PjFRLs0lotef6ueZACx2AwfTis4PDDSHxA2pOz+aOzQTFNTAUSDoQZSh3iQn8HKljWGxpT4MT5Fh9osCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cb9da6da4ca88e79a4df57960c7bd7e03848d275b8d375c633479abb351c0358","last_reissued_at":"2026-07-03T00:17:02.221940Z","signature_status":"signed_v1","first_computed_at":"2026-07-03T00:17:02.221940Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Francis Bach, Gabriel Synnaeve, Juliette Decugis, Sean O'Brien, Taco Cohen","submitted_at":"2026-07-01T21:39:19Z","abstract_excerpt":"Reinforcement learning post-training dramatically improves LLM reasoning, but suffers from training instability and diversity collapse. Advantage functions offer an appealing fix: they reshape the training objective, reweight which rollouts drive learning, and are trivial to implement. Yet a proliferation of methods makes it unclear which advantage to use and when. We cut through the confusion with a unifying framework that decomposes any advantage into its positive and negative gradient mass along two orthogonal axes. On the sign axis, imbalanced updates collapse either entropy or weight geom"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.01490","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/2607.01490/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":"2607.01490","created_at":"2026-07-03T00:17:02.222001+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.01490v1","created_at":"2026-07-03T00:17:02.222001+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.01490","created_at":"2026-07-03T00:17:02.222001+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZOO2NWSMVCHH","created_at":"2026-07-03T00:17:02.222001+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZOO2NWSMVCHHTJG7","created_at":"2026-07-03T00:17:02.222001+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZOO2NWSM","created_at":"2026-07-03T00:17:02.222001+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/ZOO2NWSMVCHHTJG7K6LAY66X4A","json":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A.json","graph_json":"https://pith.science/api/pith-number/ZOO2NWSMVCHHTJG7K6LAY66X4A/graph.json","events_json":"https://pith.science/api/pith-number/ZOO2NWSMVCHHTJG7K6LAY66X4A/events.json","paper":"https://pith.science/paper/ZOO2NWSM"},"agent_actions":{"view_html":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A","download_json":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A.json","view_paper":"https://pith.science/paper/ZOO2NWSM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.01490&json=true","fetch_graph":"https://pith.science/api/pith-number/ZOO2NWSMVCHHTJG7K6LAY66X4A/graph.json","fetch_events":"https://pith.science/api/pith-number/ZOO2NWSMVCHHTJG7K6LAY66X4A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A/action/storage_attestation","attest_author":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A/action/author_attestation","sign_citation":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A/action/citation_signature","submit_replication":"https://pith.science/pith/ZOO2NWSMVCHHTJG7K6LAY66X4A/action/replication_record"}},"created_at":"2026-07-03T00:17:02.222001+00:00","updated_at":"2026-07-03T00:17:02.222001+00:00"}