{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:A3XCM5R7IVCZOW2C77AAW3NZWD","short_pith_number":"pith:A3XCM5R7","canonical_record":{"source":{"id":"2606.31392","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-30T09:19:38Z","cross_cats_sorted":[],"title_canon_sha256":"b22402aefd588991590f907e975d25652910d3976381f6a2c3443a9475878f6a","abstract_canon_sha256":"a3f080392a98b4fadd9b931b92c88fe70d4de64553a0cf9f4162adf0efd459be"},"schema_version":"1.0"},"canonical_sha256":"06ee26763f4545975b42ffc00b6db9b0f94fe031d425109e0853a4d5053a51e8","source":{"kind":"arxiv","id":"2606.31392","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.31392","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"arxiv_version","alias_value":"2606.31392v1","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.31392","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_12","alias_value":"A3XCM5R7IVCZ","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_16","alias_value":"A3XCM5R7IVCZOW2C","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_8","alias_value":"A3XCM5R7","created_at":"2026-07-01T01:18:01Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:A3XCM5R7IVCZOW2C77AAW3NZWD","target":"record","payload":{"canonical_record":{"source":{"id":"2606.31392","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-30T09:19:38Z","cross_cats_sorted":[],"title_canon_sha256":"b22402aefd588991590f907e975d25652910d3976381f6a2c3443a9475878f6a","abstract_canon_sha256":"a3f080392a98b4fadd9b931b92c88fe70d4de64553a0cf9f4162adf0efd459be"},"schema_version":"1.0"},"canonical_sha256":"06ee26763f4545975b42ffc00b6db9b0f94fe031d425109e0853a4d5053a51e8","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-01T01:18:01.696390Z","signature_b64":"ts84qOJPzB7ahdrm/mOA2yIaHiBniHmZeVNiHZ6vQKDmrif08g37F1jAOgCapse5MvxCSnbNDMeIHZ7dBTP3Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"06ee26763f4545975b42ffc00b6db9b0f94fe031d425109e0853a4d5053a51e8","last_reissued_at":"2026-07-01T01:18:01.695899Z","signature_status":"signed_v1","first_computed_at":"2026-07-01T01:18:01.695899Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2606.31392","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:18:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KxRUzIwl15euL3BOYbPDNkUrVMzLXgxRn8P3Id0dzZy8wCyJHN4eHT9uU/i42wgm25eKkmUTGsrDY5lJUFupAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T19:14:36.438529Z"},"content_sha256":"96100a210c4c43e9061c2b7d84506365e37d3002a35a3359a2742b07c164f8b8","schema_version":"1.0","event_id":"sha256:96100a210c4c43e9061c2b7d84506365e37d3002a35a3359a2742b07c164f8b8"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:A3XCM5R7IVCZOW2C77AAW3NZWD","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"ReGRPO: Reflection-Augmented Policy Optimization for Tool-Using Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Binjie Zhang, Mike Zheng Shou","submitted_at":"2026-06-30T09:19:38Z","abstract_excerpt":"Tool-augmented vision-language models (VLMs) can solve multimodal, multi-step tasks by calling external tools, yet they remain fragile in practice. Existing works have two common gaps. Supervised fine-tuning (SFT) is built mostly on successful trajectories and offers little signal for recovery after tool failures, while sparse trajectory-level RL rewards provide limited guidance on which step failed and how to repair it. We introduce ReGRPO (Reflection-augmented Group Relative Policy Optimization), a framework that learns reflection-guided correction in tool-using agents. ReGRPO starts with a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.31392","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/2606.31392/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:18:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"QlSu7dSSgEaJqHemxRimtu/SuTKJbqbZ1njgfGpZn+2DgzRflym534nHJrGh4fOwe2NhORu1tFtdkxg3GAJhDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T19:14:36.438893Z"},"content_sha256":"8be3c4adae774f21dcb58991564f80bbfff2b90dbcd73266c658e0a1b36ed357","schema_version":"1.0","event_id":"sha256:8be3c4adae774f21dcb58991564f80bbfff2b90dbcd73266c658e0a1b36ed357"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/bundle.json","state_url":"https://pith.science/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-01T19:14:36Z","links":{"resolver":"https://pith.science/pith/A3XCM5R7IVCZOW2C77AAW3NZWD","bundle":"https://pith.science/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/bundle.json","state":"https://pith.science/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/state.json","well_known_bundle":"https://pith.science/.well-known/pith/A3XCM5R7IVCZOW2C77AAW3NZWD/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:A3XCM5R7IVCZOW2C77AAW3NZWD","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a3f080392a98b4fadd9b931b92c88fe70d4de64553a0cf9f4162adf0efd459be","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-30T09:19:38Z","title_canon_sha256":"b22402aefd588991590f907e975d25652910d3976381f6a2c3443a9475878f6a"},"schema_version":"1.0","source":{"id":"2606.31392","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.31392","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"arxiv_version","alias_value":"2606.31392v1","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.31392","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_12","alias_value":"A3XCM5R7IVCZ","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_16","alias_value":"A3XCM5R7IVCZOW2C","created_at":"2026-07-01T01:18:01Z"},{"alias_kind":"pith_short_8","alias_value":"A3XCM5R7","created_at":"2026-07-01T01:18:01Z"}],"graph_snapshots":[{"event_id":"sha256:8be3c4adae774f21dcb58991564f80bbfff2b90dbcd73266c658e0a1b36ed357","target":"graph","created_at":"2026-07-01T01:18:01Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2606.31392/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Tool-augmented vision-language models (VLMs) can solve multimodal, multi-step tasks by calling external tools, yet they remain fragile in practice. Existing works have two common gaps. Supervised fine-tuning (SFT) is built mostly on successful trajectories and offers little signal for recovery after tool failures, while sparse trajectory-level RL rewards provide limited guidance on which step failed and how to repair it. We introduce ReGRPO (Reflection-augmented Group Relative Policy Optimization), a framework that learns reflection-guided correction in tool-using agents. ReGRPO starts with a ","authors_text":"Binjie Zhang, Mike Zheng Shou","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-30T09:19:38Z","title":"ReGRPO: Reflection-Augmented Policy Optimization for Tool-Using Agents"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.31392","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:96100a210c4c43e9061c2b7d84506365e37d3002a35a3359a2742b07c164f8b8","target":"record","created_at":"2026-07-01T01:18:01Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"a3f080392a98b4fadd9b931b92c88fe70d4de64553a0cf9f4162adf0efd459be","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-30T09:19:38Z","title_canon_sha256":"b22402aefd588991590f907e975d25652910d3976381f6a2c3443a9475878f6a"},"schema_version":"1.0","source":{"id":"2606.31392","kind":"arxiv","version":1}},"canonical_sha256":"06ee26763f4545975b42ffc00b6db9b0f94fe031d425109e0853a4d5053a51e8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"06ee26763f4545975b42ffc00b6db9b0f94fe031d425109e0853a4d5053a51e8","first_computed_at":"2026-07-01T01:18:01.695899Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-01T01:18:01.695899Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ts84qOJPzB7ahdrm/mOA2yIaHiBniHmZeVNiHZ6vQKDmrif08g37F1jAOgCapse5MvxCSnbNDMeIHZ7dBTP3Bw==","signature_status":"signed_v1","signed_at":"2026-07-01T01:18:01.696390Z","signed_message":"canonical_sha256_bytes"},"source_id":"2606.31392","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:96100a210c4c43e9061c2b7d84506365e37d3002a35a3359a2742b07c164f8b8","sha256:8be3c4adae774f21dcb58991564f80bbfff2b90dbcd73266c658e0a1b36ed357"],"state_sha256":"40fb98bc8bb7ac084fbf50c186a70bdfbb4486a5dd0db7db6341a11e15383b05"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"c7UrnBO/qQ3hC+Xxv7GaURwXCe6alQNvDqc8+2cXxr9thShMcQX5Fiheq/R5yrWeBvQ9R8ppd8qH+1G94azhBw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-01T19:14:36.440793Z","bundle_sha256":"ece9a246ba224509d724e5f8b869a37a9d8da1f004609acbf824eacac875a70a"}}