{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:LQW767VY77V7TOPOFMUMGQ6ULR","short_pith_number":"pith:LQW767VY","schema_version":"1.0","canonical_sha256":"5c2dff7eb8ffebf9b9ee2b28c343d45c412813e4b5759b0bc22dbab6229eb087","source":{"kind":"arxiv","id":"2605.19447","version":1},"attestation_state":"computed","paper":{"title":"What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dahua Lin, Kai Chen, Linyang Li, Peiji Li, Qipeng Guo, Tianyi Lyu, Xiaozhe Li, Yang Li, Yichuan Ma","submitted_at":"2026-05-19T07:00:55Z","abstract_excerpt":"Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignment remains challenging: a single success-or-failure signal must be distributed across many actions. Existing methods rely on trajectory-level rewards or proxy signals, without fully leveraging per-step environmental feedback. Multi-turn agent settings are underexplored, where feedback can include error messages, page changes, observations, or reference trajectories. We systematically study five feedback sources and two insertion granularities and introduce SERL, a selective environment-reweigh"},"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":"2605.19447","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-19T07:00:55Z","cross_cats_sorted":[],"title_canon_sha256":"064fc313e376ab50a2074477019b163f6c1986810ed2e6282a0f9b9597ff7d35","abstract_canon_sha256":"c0c21e016c52089a2a56695542990c68781619d23f52ae58a2d315feebab34df"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:46.354417Z","signature_b64":"+VTT35KK2sGAOXsQEWsexr+4WXEtzwxXbS8SXHf9ku6Q997Z4z2Hhto9oV+1BNuFFQ5e92B7oK7565BGDjeADw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5c2dff7eb8ffebf9b9ee2b28c343d45c412813e4b5759b0bc22dbab6229eb087","last_reissued_at":"2026-05-20T01:05:46.353563Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:46.353563Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Dahua Lin, Kai Chen, Linyang Li, Peiji Li, Qipeng Guo, Tianyi Lyu, Xiaozhe Li, Yang Li, Yichuan Ma","submitted_at":"2026-05-19T07:00:55Z","abstract_excerpt":"Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignment remains challenging: a single success-or-failure signal must be distributed across many actions. Existing methods rely on trajectory-level rewards or proxy signals, without fully leveraging per-step environmental feedback. Multi-turn agent settings are underexplored, where feedback can include error messages, page changes, observations, or reference trajectories. We systematically study five feedback sources and two insertion granularities and introduce SERL, a selective environment-reweigh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.19447","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/2605.19447/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":"2605.19447","created_at":"2026-05-20T01:05:46.353707+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.19447v1","created_at":"2026-05-20T01:05:46.353707+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.19447","created_at":"2026-05-20T01:05:46.353707+00:00"},{"alias_kind":"pith_short_12","alias_value":"LQW767VY77V7","created_at":"2026-05-20T01:05:46.353707+00:00"},{"alias_kind":"pith_short_16","alias_value":"LQW767VY77V7TOPO","created_at":"2026-05-20T01:05:46.353707+00:00"},{"alias_kind":"pith_short_8","alias_value":"LQW767VY","created_at":"2026-05-20T01:05:46.353707+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/LQW767VY77V7TOPOFMUMGQ6ULR","json":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR.json","graph_json":"https://pith.science/api/pith-number/LQW767VY77V7TOPOFMUMGQ6ULR/graph.json","events_json":"https://pith.science/api/pith-number/LQW767VY77V7TOPOFMUMGQ6ULR/events.json","paper":"https://pith.science/paper/LQW767VY"},"agent_actions":{"view_html":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR","download_json":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR.json","view_paper":"https://pith.science/paper/LQW767VY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.19447&json=true","fetch_graph":"https://pith.science/api/pith-number/LQW767VY77V7TOPOFMUMGQ6ULR/graph.json","fetch_events":"https://pith.science/api/pith-number/LQW767VY77V7TOPOFMUMGQ6ULR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR/action/storage_attestation","attest_author":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR/action/author_attestation","sign_citation":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR/action/citation_signature","submit_replication":"https://pith.science/pith/LQW767VY77V7TOPOFMUMGQ6ULR/action/replication_record"}},"created_at":"2026-05-20T01:05:46.353707+00:00","updated_at":"2026-05-20T01:05:46.353707+00:00"}