{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:5VIYF53XPXTXWRQRY3ER23MPDI","short_pith_number":"pith:5VIYF53X","canonical_record":{"source":{"id":"2607.24484","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-27T14:20:59Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"3d34eb2e510b1a096aae48d616484002256fa60697dd43cecacbd8cd7473f943","abstract_canon_sha256":"cd2d0192b11d0d4c4c5f5b222901a0d3bdedba8ccdd6574459fbc2f656b1779f"},"schema_version":"1.0"},"canonical_sha256":"ed5182f7777de77b4611c6c91d6d8f1a1f59563a141cfed314f934a3cb7c654e","source":{"kind":"arxiv","id":"2607.24484","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.24484","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"arxiv_version","alias_value":"2607.24484v1","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.24484","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_12","alias_value":"5VIYF53XPXTX","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_16","alias_value":"5VIYF53XPXTXWRQR","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_8","alias_value":"5VIYF53X","created_at":"2026-07-28T02:24:06Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:5VIYF53XPXTXWRQRY3ER23MPDI","target":"record","payload":{"canonical_record":{"source":{"id":"2607.24484","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-27T14:20:59Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"3d34eb2e510b1a096aae48d616484002256fa60697dd43cecacbd8cd7473f943","abstract_canon_sha256":"cd2d0192b11d0d4c4c5f5b222901a0d3bdedba8ccdd6574459fbc2f656b1779f"},"schema_version":"1.0"},"canonical_sha256":"ed5182f7777de77b4611c6c91d6d8f1a1f59563a141cfed314f934a3cb7c654e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T02:24:06.068117Z","signature_b64":"WVn5wSAXARPccnuv3z8tvLij/V3aaycvzVIJbOLECgnPINz2SU6mlji9JrIVPtGidvQLakUkr4FSx62z/5lpBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ed5182f7777de77b4611c6c91d6d8f1a1f59563a141cfed314f934a3cb7c654e","last_reissued_at":"2026-07-28T02:24:06.067216Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T02:24:06.067216Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2607.24484","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":"2026-07-28T02:24:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"YqDk93A+hrGvFrkNjUUD36pegxdGf0IIEvacSaY/ztWJsnuLmUjZb6Uiecw1M6mwGRHXdwtn5j+R/+ofRpCEAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-01T22:50:00.753722Z"},"content_sha256":"6bbaec484460838eb83ce11ff364f49de65a0b52e5d1ffbd9677c843e18cc20d","schema_version":"1.0","event_id":"sha256:6bbaec484460838eb83ce11ff364f49de65a0b52e5d1ffbd9677c843e18cc20d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:5VIYF53XPXTXWRQRY3ER23MPDI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"What do Reward Models Memorize?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ekaterina Shutova, Ivo Verhoeven, Pushkar Mishra","submitted_at":"2026-07-27T14:20:59Z","abstract_excerpt":"This paper studies what discriminatively trained reward models (RMs) memorize by measuring counterfactual memorization on two human preference datasets. We show that RMs 1) misallocate memorization to easy, high margin preference pairs, 2) memorize dataset-specific shortcuts (e.g., model identity, user sampling strategy), and 3) overgeneralize simple heuristic correlates of human preference (e.g., length, compliance) when confronted with unseen preference pairs. Overall, our findings indicate that discriminative training of RMs from human preference data results in biased RMs not yet capable o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.24484","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/2607.24484/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":"2026-07-28T02:24:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pkgaxVC5rUYpTErZRDS2m8Ui6pYCZIGi1KBMfk72t3PQBhQzmD/Z3Wr9cPzloEQzacyoyo7r/Zkm757UWEZyDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-01T22:50:00.754407Z"},"content_sha256":"c317ff6f45ab68299860601d28f37df08e9dfc463736b3dc19a258f0c825547c","schema_version":"1.0","event_id":"sha256:c317ff6f45ab68299860601d28f37df08e9dfc463736b3dc19a258f0c825547c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/5VIYF53XPXTXWRQRY3ER23MPDI/bundle.json","state_url":"https://pith.science/pith/5VIYF53XPXTXWRQRY3ER23MPDI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/5VIYF53XPXTXWRQRY3ER23MPDI/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":"2026-08-01T22:50:00Z","links":{"resolver":"https://pith.science/pith/5VIYF53XPXTXWRQRY3ER23MPDI","bundle":"https://pith.science/pith/5VIYF53XPXTXWRQRY3ER23MPDI/bundle.json","state":"https://pith.science/pith/5VIYF53XPXTXWRQRY3ER23MPDI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/5VIYF53XPXTXWRQRY3ER23MPDI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:5VIYF53XPXTXWRQRY3ER23MPDI","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":"cd2d0192b11d0d4c4c5f5b222901a0d3bdedba8ccdd6574459fbc2f656b1779f","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-27T14:20:59Z","title_canon_sha256":"3d34eb2e510b1a096aae48d616484002256fa60697dd43cecacbd8cd7473f943"},"schema_version":"1.0","source":{"id":"2607.24484","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.24484","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"arxiv_version","alias_value":"2607.24484v1","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.24484","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_12","alias_value":"5VIYF53XPXTX","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_16","alias_value":"5VIYF53XPXTXWRQR","created_at":"2026-07-28T02:24:06Z"},{"alias_kind":"pith_short_8","alias_value":"5VIYF53X","created_at":"2026-07-28T02:24:06Z"}],"graph_snapshots":[{"event_id":"sha256:c317ff6f45ab68299860601d28f37df08e9dfc463736b3dc19a258f0c825547c","target":"graph","created_at":"2026-07-28T02:24:06Z","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/2607.24484/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"This paper studies what discriminatively trained reward models (RMs) memorize by measuring counterfactual memorization on two human preference datasets. We show that RMs 1) misallocate memorization to easy, high margin preference pairs, 2) memorize dataset-specific shortcuts (e.g., model identity, user sampling strategy), and 3) overgeneralize simple heuristic correlates of human preference (e.g., length, compliance) when confronted with unseen preference pairs. Overall, our findings indicate that discriminative training of RMs from human preference data results in biased RMs not yet capable o","authors_text":"Ekaterina Shutova, Ivo Verhoeven, Pushkar Mishra","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-27T14:20:59Z","title":"What do Reward Models Memorize?"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.24484","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":"sha256:6bbaec484460838eb83ce11ff364f49de65a0b52e5d1ffbd9677c843e18cc20d","target":"record","created_at":"2026-07-28T02:24:06Z","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":"cd2d0192b11d0d4c4c5f5b222901a0d3bdedba8ccdd6574459fbc2f656b1779f","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-27T14:20:59Z","title_canon_sha256":"3d34eb2e510b1a096aae48d616484002256fa60697dd43cecacbd8cd7473f943"},"schema_version":"1.0","source":{"id":"2607.24484","kind":"arxiv","version":1}},"canonical_sha256":"ed5182f7777de77b4611c6c91d6d8f1a1f59563a141cfed314f934a3cb7c654e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ed5182f7777de77b4611c6c91d6d8f1a1f59563a141cfed314f934a3cb7c654e","first_computed_at":"2026-07-28T02:24:06.067216Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-28T02:24:06.067216Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"WVn5wSAXARPccnuv3z8tvLij/V3aaycvzVIJbOLECgnPINz2SU6mlji9JrIVPtGidvQLakUkr4FSx62z/5lpBw==","signature_status":"signed_v1","signed_at":"2026-07-28T02:24:06.068117Z","signed_message":"canonical_sha256_bytes"},"source_id":"2607.24484","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6bbaec484460838eb83ce11ff364f49de65a0b52e5d1ffbd9677c843e18cc20d","sha256:c317ff6f45ab68299860601d28f37df08e9dfc463736b3dc19a258f0c825547c"],"state_sha256":"86f807d4e1f09a8c308a3cbfc62c46abfba1c8562176c628c5de5a4ece3252a0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DMfK5P6xf6Uby4MOHyqK4Qh6jqGog6bha7NMEUBKmjHeOZ2LV1n2vGObcRDTVgDxIVEMMdinGA6dABT+4pLMAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-01T22:50:00.760335Z","bundle_sha256":"129974ad2a45845103b40852e4987675dcdcac93d53941f38922d9b7fb3ef774"}}