{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:YG52KFSTNA5AOZRNZQEI5ZLJKG","short_pith_number":"pith:YG52KFST","canonical_record":{"source":{"id":"2606.03238","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T06:55:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"fb253d5a0f3264a8eebb6ea8cc38257b2defdcd7c246c1b70fa652c03d43dbea","abstract_canon_sha256":"c4c2f92a92b99d2ea2daba9c8211f4d513c2915b8d64381c2144e63e7b901a36"},"schema_version":"1.0"},"canonical_sha256":"c1bba51653683a07662dcc088ee56951b07663196fefca7a962d3fb5a16220df","source":{"kind":"arxiv","id":"2606.03238","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.03238","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"arxiv_version","alias_value":"2606.03238v1","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.03238","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_12","alias_value":"YG52KFSTNA5A","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_16","alias_value":"YG52KFSTNA5AOZRN","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_8","alias_value":"YG52KFST","created_at":"2026-06-03T01:05:35Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:YG52KFSTNA5AOZRNZQEI5ZLJKG","target":"record","payload":{"canonical_record":{"source":{"id":"2606.03238","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T06:55:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"fb253d5a0f3264a8eebb6ea8cc38257b2defdcd7c246c1b70fa652c03d43dbea","abstract_canon_sha256":"c4c2f92a92b99d2ea2daba9c8211f4d513c2915b8d64381c2144e63e7b901a36"},"schema_version":"1.0"},"canonical_sha256":"c1bba51653683a07662dcc088ee56951b07663196fefca7a962d3fb5a16220df","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-03T01:05:35.725070Z","signature_b64":"HOUctYLWhSRy5facdDX+4N9+1TGqT3c8r8/cOzDXwBiAeIsYwxPRBP5R3hWDBd+Rm/a8+BUXjYmbp8BFWenBDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1bba51653683a07662dcc088ee56951b07663196fefca7a962d3fb5a16220df","last_reissued_at":"2026-06-03T01:05:35.724575Z","signature_status":"signed_v1","first_computed_at":"2026-06-03T01:05:35.724575Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2606.03238","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":"2026-06-03T01:05:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"zvmGuJjpkot/hW7le28NRC8USKgUq/thLdD//7XDon/wDD5cD9UqKbFQERSmWOAuCVuhF1qmNMmKEtwHV2YoBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-05T01:42:43.689947Z"},"content_sha256":"bbd1c73bb91eaf529690443cc68f4d0ebf12bd96aaf312e7495d9293e26c1c21","schema_version":"1.0","event_id":"sha256:bbd1c73bb91eaf529690443cc68f4d0ebf12bd96aaf312e7495d9293e26c1c21"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:YG52KFSTNA5AOZRNZQEI5ZLJKG","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"When RLHF Fails: A Mechanistic Taxonomy of Reward Hacking, Collapse, and Evaluator Gaming","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Zelalem Abahana","submitted_at":"2026-06-02T06:55:52Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) makes large-scale post-training possible by replacing an underspecified human objective with learned and scalable proxies. The same substitution creates a structured failure surface: optimization can raise the learned reward while external quality falls, degrade both proxy and judge scores, reveal proxy under-alignment, or produce evaluator-specific disagreement. We present an empirical failure-mode study of a compact RLHF pipeline with proximal policy optimization (PPO), direct preference optimization (DPO), uncertainty-penalized PPO (UP-PPO),"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.03238","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/2606.03238/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":"2026-06-03T01:05:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4rf1E4Ibhpn+7H3zd+RKy44twPcz6yiQ9udDzeX/hst/88lp8H4XK9HocE821aVzdlS/3iHN6OR46tJcazHiAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-05T01:42:43.690337Z"},"content_sha256":"addb4a3c876228de53fc5ac345421efc5ed50a0fc81f164b086308d7f4d71db8","schema_version":"1.0","event_id":"sha256:addb4a3c876228de53fc5ac345421efc5ed50a0fc81f164b086308d7f4d71db8"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/bundle.json","state_url":"https://pith.science/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/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":"2026-07-05T01:42:43Z","links":{"resolver":"https://pith.science/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG","bundle":"https://pith.science/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/bundle.json","state":"https://pith.science/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/state.json","well_known_bundle":"https://pith.science/.well-known/pith/YG52KFSTNA5AOZRNZQEI5ZLJKG/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:YG52KFSTNA5AOZRNZQEI5ZLJKG","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":"c4c2f92a92b99d2ea2daba9c8211f4d513c2915b8d64381c2144e63e7b901a36","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T06:55:52Z","title_canon_sha256":"fb253d5a0f3264a8eebb6ea8cc38257b2defdcd7c246c1b70fa652c03d43dbea"},"schema_version":"1.0","source":{"id":"2606.03238","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.03238","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"arxiv_version","alias_value":"2606.03238v1","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.03238","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_12","alias_value":"YG52KFSTNA5A","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_16","alias_value":"YG52KFSTNA5AOZRN","created_at":"2026-06-03T01:05:35Z"},{"alias_kind":"pith_short_8","alias_value":"YG52KFST","created_at":"2026-06-03T01:05:35Z"}],"graph_snapshots":[{"event_id":"sha256:addb4a3c876228de53fc5ac345421efc5ed50a0fc81f164b086308d7f4d71db8","target":"graph","created_at":"2026-06-03T01:05:35Z","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/2606.03238/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement learning from human feedback (RLHF) makes large-scale post-training possible by replacing an underspecified human objective with learned and scalable proxies. The same substitution creates a structured failure surface: optimization can raise the learned reward while external quality falls, degrade both proxy and judge scores, reveal proxy under-alignment, or produce evaluator-specific disagreement. We present an empirical failure-mode study of a compact RLHF pipeline with proximal policy optimization (PPO), direct preference optimization (DPO), uncertainty-penalized PPO (UP-PPO),","authors_text":"Zelalem Abahana","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T06:55:52Z","title":"When RLHF Fails: A Mechanistic Taxonomy of Reward Hacking, Collapse, and Evaluator Gaming"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.03238","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:bbd1c73bb91eaf529690443cc68f4d0ebf12bd96aaf312e7495d9293e26c1c21","target":"record","created_at":"2026-06-03T01:05:35Z","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":"c4c2f92a92b99d2ea2daba9c8211f4d513c2915b8d64381c2144e63e7b901a36","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T06:55:52Z","title_canon_sha256":"fb253d5a0f3264a8eebb6ea8cc38257b2defdcd7c246c1b70fa652c03d43dbea"},"schema_version":"1.0","source":{"id":"2606.03238","kind":"arxiv","version":1}},"canonical_sha256":"c1bba51653683a07662dcc088ee56951b07663196fefca7a962d3fb5a16220df","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c1bba51653683a07662dcc088ee56951b07663196fefca7a962d3fb5a16220df","first_computed_at":"2026-06-03T01:05:35.724575Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-03T01:05:35.724575Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HOUctYLWhSRy5facdDX+4N9+1TGqT3c8r8/cOzDXwBiAeIsYwxPRBP5R3hWDBd+Rm/a8+BUXjYmbp8BFWenBDw==","signature_status":"signed_v1","signed_at":"2026-06-03T01:05:35.725070Z","signed_message":"canonical_sha256_bytes"},"source_id":"2606.03238","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:bbd1c73bb91eaf529690443cc68f4d0ebf12bd96aaf312e7495d9293e26c1c21","sha256:addb4a3c876228de53fc5ac345421efc5ed50a0fc81f164b086308d7f4d71db8"],"state_sha256":"50dcbbe4542530b50c3f72c817e91e587871eb7ba45a0887873b44f8bb696397"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JatDBIpyTKcjKVBkftB0focYDOc5Ue22Gpw/u2QzQ4GZX7z7eZ8LfLdTxFsKQW/5MTx9Iol8zTFEoL//9U8LDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-05T01:42:43.692258Z","bundle_sha256":"0419aab87b08a996c3f89670651a53899d84220f6d56e1a0141e80328d94d867"}}