{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GYXRWSMJUUAG5PT5GSC4W2ZQG2","short_pith_number":"pith:GYXRWSMJ","schema_version":"1.0","canonical_sha256":"362f1b4989a5006ebe7d3485cb6b3036b7d6140c59c5746def047fd20a8a18d6","source":{"kind":"arxiv","id":"2607.27787","version":1},"attestation_state":"computed","paper":{"title":"LoRA Scaffolded Policy Optimization (LSPO): A Sampling-Time Low-Rank Scaffold for Recovering Reinforcement-Learning Gradient on Zero-Reward Cliff Prompts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ken Ding","submitted_at":"2026-07-30T07:21:34Z","abstract_excerpt":"Reinforcement learning from verifiable rewards (RLVR) for mathematical reasoning suffers from a structural blind spot: on \"cliff\" prompts-those on which every sampled rollout in a group fails-the group-normalized advantage is identically zero, so GRPO produces no gradient on precisely the prompts at the frontier of the model's capability. We introduce LoRA Scaffolded Policy Optimization (LSPO), a sampling-time mechanism that recovers this lost gradient. Each RL step, LSPO detects cliff prompts, fits a small low-rank (LoRA) adapter by a brief supervised step on their ground-truth solutions, re-"},"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":"2607.27787","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-30T07:21:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"63695f9a24abcacaf3b088ffc8da7302f2c99f770fda9e9ca152ca7c8ab4cc1b","abstract_canon_sha256":"f09e06b9ad0b6ab2481dac3c09ebec1664b3836fc305fe877800d42f8212bea7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"362f1b4989a5006ebe7d3485cb6b3036b7d6140c59c5746def047fd20a8a18d6","last_reissued_at":"2026-07-31T01:30:47.366553Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-31T01:30:47.366553Z"},"graph_snapshot":{"paper":{"title":"LoRA Scaffolded Policy Optimization (LSPO): A Sampling-Time Low-Rank Scaffold for Recovering Reinforcement-Learning Gradient on Zero-Reward Cliff Prompts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ken Ding","submitted_at":"2026-07-30T07:21:34Z","abstract_excerpt":"Reinforcement learning from verifiable rewards (RLVR) for mathematical reasoning suffers from a structural blind spot: on \"cliff\" prompts-those on which every sampled rollout in a group fails-the group-normalized advantage is identically zero, so GRPO produces no gradient on precisely the prompts at the frontier of the model's capability. We introduce LoRA Scaffolded Policy Optimization (LSPO), a sampling-time mechanism that recovers this lost gradient. Each RL step, LSPO detects cliff prompts, fits a small low-rank (LoRA) adapter by a brief supervised step on their ground-truth solutions, re-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.27787","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/2607.27787/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":"2607.27787","created_at":"2026-07-31T01:30:47.369710+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.27787v1","created_at":"2026-07-31T01:30:47.369710+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.27787","created_at":"2026-07-31T01:30:47.369710+00:00"},{"alias_kind":"pith_short_12","alias_value":"GYXRWSMJUUAG","created_at":"2026-07-31T01:30:47.369710+00:00"},{"alias_kind":"pith_short_16","alias_value":"GYXRWSMJUUAG5PT5","created_at":"2026-07-31T01:30:47.369710+00:00"},{"alias_kind":"pith_short_8","alias_value":"GYXRWSMJ","created_at":"2026-07-31T01:30:47.369710+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/GYXRWSMJUUAG5PT5GSC4W2ZQG2","json":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2.json","graph_json":"https://pith.science/api/pith-number/GYXRWSMJUUAG5PT5GSC4W2ZQG2/graph.json","events_json":"https://pith.science/api/pith-number/GYXRWSMJUUAG5PT5GSC4W2ZQG2/events.json","paper":"https://pith.science/paper/GYXRWSMJ"},"agent_actions":{"view_html":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2","download_json":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2.json","view_paper":"https://pith.science/paper/GYXRWSMJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.27787&json=true","fetch_graph":"https://pith.science/api/pith-number/GYXRWSMJUUAG5PT5GSC4W2ZQG2/graph.json","fetch_events":"https://pith.science/api/pith-number/GYXRWSMJUUAG5PT5GSC4W2ZQG2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2/action/storage_attestation","attest_author":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2/action/author_attestation","sign_citation":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2/action/citation_signature","submit_replication":"https://pith.science/pith/GYXRWSMJUUAG5PT5GSC4W2ZQG2/action/replication_record"}},"created_at":"2026-07-31T01:30:47.369710+00:00","updated_at":"2026-07-31T01:30:47.369710+00:00"}