{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:EOQZGMUF5W65JSQQPAUGZJNNGA","short_pith_number":"pith:EOQZGMUF","canonical_record":{"source":{"id":"2506.02355","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T01:15:15Z","cross_cats_sorted":[],"title_canon_sha256":"63c9a06f85bf570ac08526abbc4d78f5949a73adb832f290582f085586f6d4f0","abstract_canon_sha256":"4b508de4a02db58e113f64fdce8377b03a2e234e3535af5ea3f65e81ce57d59a"},"schema_version":"1.0"},"canonical_sha256":"23a1933285edbdd4ca1078286ca5ad300b35bea4a2b6167993bff0468c197140","source":{"kind":"arxiv","id":"2506.02355","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.02355","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"arxiv_version","alias_value":"2506.02355v2","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02355","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_12","alias_value":"EOQZGMUF5W65","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_16","alias_value":"EOQZGMUF5W65JSQQ","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_8","alias_value":"EOQZGMUF","created_at":"2026-07-05T11:24:21Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:EOQZGMUF5W65JSQQPAUGZJNNGA","target":"record","payload":{"canonical_record":{"source":{"id":"2506.02355","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T01:15:15Z","cross_cats_sorted":[],"title_canon_sha256":"63c9a06f85bf570ac08526abbc4d78f5949a73adb832f290582f085586f6d4f0","abstract_canon_sha256":"4b508de4a02db58e113f64fdce8377b03a2e234e3535af5ea3f65e81ce57d59a"},"schema_version":"1.0"},"canonical_sha256":"23a1933285edbdd4ca1078286ca5ad300b35bea4a2b6167993bff0468c197140","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:24:21.784251Z","signature_b64":"5tUI0J/RVzqk9MQ1e5i0o62u4HU8eSyHoeYpOJzSbtVpFgl7uW8PbHZIibVSE1YoXfVsH3cypUC6P3ZgjAlQDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"23a1933285edbdd4ca1078286ca5ad300b35bea4a2b6167993bff0468c197140","last_reissued_at":"2026-07-05T11:24:21.783767Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:24:21.783767Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2506.02355","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:24:21Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"QJgVmC+0l9PGn2abl5KievD3zglH2lACW4ev69kwfWMIB1FcjQW/XjEwfUtg1KbwPG5LB8HZp8xFdEcd0x40AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-06T08:53:56.025164Z"},"content_sha256":"fb38de1b593b403aaf9a71df523c558b3432dfd50b1af321b5daaf973cb8b96d","schema_version":"1.0","event_id":"sha256:fb38de1b593b403aaf9a71df523c558b3432dfd50b1af321b5daaf973cb8b96d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:EOQZGMUF5W65JSQQPAUGZJNNGA","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Rewarding the Unlikely: Lifting GRPO Beyond Distribution Sharpening","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andre He, Daniel Fried, Sean Welleck","submitted_at":"2025-06-03T01:15:15Z","abstract_excerpt":"Reinforcement learning is emerging as a primary driver for improving language model reasoning capabilities. A fundamental question is whether current reinforcement learning algorithms -- such as Group Relative Policy Optimization (GRPO), the de facto standard algorithm used to improve language model reasoning -- merely sharpen the base model's distribution around problems it can already solve. We investigate this question in the context of formal theorem proving, which has access to a perfect verifier. We identify a degenerate rank bias in GRPO in which highly probable trajectories are reinfor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02355","kind":"arxiv","version":2},"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/2506.02355/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":"2026-07-05T11:24:21Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"XozH1Hy2uj3/vlA738a0VESSsafIwnZTFT2FI+3zEkp06x8sNdp4z7rrCqzRk9bmndeKtqjYi2nNsZy5bpflAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-06T08:53:56.025553Z"},"content_sha256":"20f3054a30433f2e2cf05e28af7a7ecb5db1a3335ae56d2556f86757e54d4a4c","schema_version":"1.0","event_id":"sha256:20f3054a30433f2e2cf05e28af7a7ecb5db1a3335ae56d2556f86757e54d4a4c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/bundle.json","state_url":"https://pith.science/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/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":"2026-07-06T08:53:56Z","links":{"resolver":"https://pith.science/pith/EOQZGMUF5W65JSQQPAUGZJNNGA","bundle":"https://pith.science/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/bundle.json","state":"https://pith.science/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/state.json","well_known_bundle":"https://pith.science/.well-known/pith/EOQZGMUF5W65JSQQPAUGZJNNGA/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:EOQZGMUF5W65JSQQPAUGZJNNGA","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":"4b508de4a02db58e113f64fdce8377b03a2e234e3535af5ea3f65e81ce57d59a","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T01:15:15Z","title_canon_sha256":"63c9a06f85bf570ac08526abbc4d78f5949a73adb832f290582f085586f6d4f0"},"schema_version":"1.0","source":{"id":"2506.02355","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.02355","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"arxiv_version","alias_value":"2506.02355v2","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02355","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_12","alias_value":"EOQZGMUF5W65","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_16","alias_value":"EOQZGMUF5W65JSQQ","created_at":"2026-07-05T11:24:21Z"},{"alias_kind":"pith_short_8","alias_value":"EOQZGMUF","created_at":"2026-07-05T11:24:21Z"}],"graph_snapshots":[{"event_id":"sha256:20f3054a30433f2e2cf05e28af7a7ecb5db1a3335ae56d2556f86757e54d4a4c","target":"graph","created_at":"2026-07-05T11:24:21Z","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/2506.02355/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement learning is emerging as a primary driver for improving language model reasoning capabilities. A fundamental question is whether current reinforcement learning algorithms -- such as Group Relative Policy Optimization (GRPO), the de facto standard algorithm used to improve language model reasoning -- merely sharpen the base model's distribution around problems it can already solve. We investigate this question in the context of formal theorem proving, which has access to a perfect verifier. We identify a degenerate rank bias in GRPO in which highly probable trajectories are reinfor","authors_text":"Andre He, Daniel Fried, Sean Welleck","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T01:15:15Z","title":"Rewarding the Unlikely: Lifting GRPO Beyond Distribution Sharpening"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02355","kind":"arxiv","version":2},"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:fb38de1b593b403aaf9a71df523c558b3432dfd50b1af321b5daaf973cb8b96d","target":"record","created_at":"2026-07-05T11:24:21Z","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":"4b508de4a02db58e113f64fdce8377b03a2e234e3535af5ea3f65e81ce57d59a","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T01:15:15Z","title_canon_sha256":"63c9a06f85bf570ac08526abbc4d78f5949a73adb832f290582f085586f6d4f0"},"schema_version":"1.0","source":{"id":"2506.02355","kind":"arxiv","version":2}},"canonical_sha256":"23a1933285edbdd4ca1078286ca5ad300b35bea4a2b6167993bff0468c197140","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"23a1933285edbdd4ca1078286ca5ad300b35bea4a2b6167993bff0468c197140","first_computed_at":"2026-07-05T11:24:21.783767Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:24:21.783767Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"5tUI0J/RVzqk9MQ1e5i0o62u4HU8eSyHoeYpOJzSbtVpFgl7uW8PbHZIibVSE1YoXfVsH3cypUC6P3ZgjAlQDw==","signature_status":"signed_v1","signed_at":"2026-07-05T11:24:21.784251Z","signed_message":"canonical_sha256_bytes"},"source_id":"2506.02355","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:fb38de1b593b403aaf9a71df523c558b3432dfd50b1af321b5daaf973cb8b96d","sha256:20f3054a30433f2e2cf05e28af7a7ecb5db1a3335ae56d2556f86757e54d4a4c"],"state_sha256":"4b3860d0f127587fab48756b717250683a4fb3e354719aa9c0e26865432635b9"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HiplYQLwNK3td0p3aNkhpWIL1E993OghIAth9EIX/e/XS/LTKBwfIuiNL7RigISGXy0ph+4UG7LF38qzmn4ICQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-06T08:53:56.027570Z","bundle_sha256":"0bcd63b94aa903251a84b2b59195fe15f6a1d5cc4858debcb2cb014a4a637837"}}