{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:D5HYECPFDLFK7TLTUIU4TOT6TE","short_pith_number":"pith:D5HYECPF","schema_version":"1.0","canonical_sha256":"1f4f8209e51acaafcd73a229c9ba7e9931539f4f63aeb4958d96de4319f6c34c","source":{"kind":"arxiv","id":"2607.10848","version":1},"attestation_state":"computed","paper":{"title":"Predictive Divergence Masks for LLM RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bowen Ping, Haonan Wang, Jiaqi Tang, Jiarui Yao, Penghui Qi, Tianyu Pang, Xiangxin Zhou","submitted_at":"2026-07-12T17:20:44Z","abstract_excerpt":"Reinforcement learning for large language models (LLMs) typically relies on trust-region masks to stabilize off-policy updates. The dominant PPO-style approach uses the sampled-token importance ratio for two criteria: a proximity criterion, which asks whether the policy has moved too far from the behavior policy, and a direction criterion, which asks whether the update pushes it farther away. Recent work DPPO improves the proximity criterion by replacing PPO's ratio-based test with a probability divergence between the behavior and training policies. However, its direction criterion is still in"},"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":"2607.10848","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-12T17:20:44Z","cross_cats_sorted":[],"title_canon_sha256":"bc1ab58e002173604a716a7e3a3961626b276bf951b6477b923cf6f26fe245c3","abstract_canon_sha256":"36e9516c3bc5abb8013ec7978868f464b6dc5df965e17c3be9938f55f06453ce"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-14T01:21:43.400579Z","signature_b64":"AMc73wTLtlxkfL0KhJT/GSI6+4j2DDjLhbAfoZzqQ0woMK+vozrPZQ8vugX93lNyBQOTSwg/D7/znH2QRSUSCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1f4f8209e51acaafcd73a229c9ba7e9931539f4f63aeb4958d96de4319f6c34c","last_reissued_at":"2026-07-14T01:21:43.399699Z","signature_status":"signed_v1","first_computed_at":"2026-07-14T01:21:43.399699Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Predictive Divergence Masks for LLM RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bowen Ping, Haonan Wang, Jiaqi Tang, Jiarui Yao, Penghui Qi, Tianyu Pang, Xiangxin Zhou","submitted_at":"2026-07-12T17:20:44Z","abstract_excerpt":"Reinforcement learning for large language models (LLMs) typically relies on trust-region masks to stabilize off-policy updates. The dominant PPO-style approach uses the sampled-token importance ratio for two criteria: a proximity criterion, which asks whether the policy has moved too far from the behavior policy, and a direction criterion, which asks whether the update pushes it farther away. Recent work DPPO improves the proximity criterion by replacing PPO's ratio-based test with a probability divergence between the behavior and training policies. However, its direction criterion is still in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.10848","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/2607.10848/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":"2607.10848","created_at":"2026-07-14T01:21:43.400157+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.10848v1","created_at":"2026-07-14T01:21:43.400157+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.10848","created_at":"2026-07-14T01:21:43.400157+00:00"},{"alias_kind":"pith_short_12","alias_value":"D5HYECPFDLFK","created_at":"2026-07-14T01:21:43.400157+00:00"},{"alias_kind":"pith_short_16","alias_value":"D5HYECPFDLFK7TLT","created_at":"2026-07-14T01:21:43.400157+00:00"},{"alias_kind":"pith_short_8","alias_value":"D5HYECPF","created_at":"2026-07-14T01:21:43.400157+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/D5HYECPFDLFK7TLTUIU4TOT6TE","json":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE.json","graph_json":"https://pith.science/api/pith-number/D5HYECPFDLFK7TLTUIU4TOT6TE/graph.json","events_json":"https://pith.science/api/pith-number/D5HYECPFDLFK7TLTUIU4TOT6TE/events.json","paper":"https://pith.science/paper/D5HYECPF"},"agent_actions":{"view_html":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE","download_json":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE.json","view_paper":"https://pith.science/paper/D5HYECPF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.10848&json=true","fetch_graph":"https://pith.science/api/pith-number/D5HYECPFDLFK7TLTUIU4TOT6TE/graph.json","fetch_events":"https://pith.science/api/pith-number/D5HYECPFDLFK7TLTUIU4TOT6TE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE/action/storage_attestation","attest_author":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE/action/author_attestation","sign_citation":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE/action/citation_signature","submit_replication":"https://pith.science/pith/D5HYECPFDLFK7TLTUIU4TOT6TE/action/replication_record"}},"created_at":"2026-07-14T01:21:43.400157+00:00","updated_at":"2026-07-14T01:21:43.400157+00:00"}