{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VTEXE334DGF3QC6EIU5TXOHBJE","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":"e83e96a90d1c1027b3412052968f37772dc914a154b20ad7d4f86d123f5ab6db","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-15T01:57:11Z","title_canon_sha256":"23a708fc428a0e28c2bf7fc6b9c21bee42b41a0f16031c2e27dbb82b71c0b6e5"},"schema_version":"1.0","source":{"id":"2605.15529","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.15529","created_at":"2026-05-20T00:01:03Z"},{"alias_kind":"arxiv_version","alias_value":"2605.15529v1","created_at":"2026-05-20T00:01:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15529","created_at":"2026-05-20T00:01:03Z"},{"alias_kind":"pith_short_12","alias_value":"VTEXE334DGF3","created_at":"2026-05-20T00:01:03Z"},{"alias_kind":"pith_short_16","alias_value":"VTEXE334DGF3QC6E","created_at":"2026-05-20T00:01:03Z"},{"alias_kind":"pith_short_8","alias_value":"VTEXE334","created_at":"2026-05-20T00:01:03Z"}],"graph_snapshots":[{"event_id":"sha256:6fa2ed170304c60bd942a6fef565058c0065c4ddde03b75748d19dcd7246b0e1","target":"graph","created_at":"2026-05-20T00:01:03Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Experiments across four backbones and four reasoning benchmarks show that BetaPRM improves PRM-guided Best-of-N selection while preserving standard step-level error detection. Built on this signal, ACA improves the accuracy--token tradeoff over fixed-budget Best-of-16, reducing token usage by up to 33.57% while improving final-answer accuracy."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that the Beta posterior learned from finite Monte Carlo continuations provides a calibrated and actionable measure of true prediction reliability rather than merely reflecting sampling noise or model bias in the continuation process itself (implicit in the Beta-Binomial likelihood description)."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"BetaPRM learns distributional step rewards with explicit reliability via Beta-Binomial modeling, enabling ACA that cuts token use by up to 33.57% while raising final-answer accuracy on reasoning benchmarks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search."}],"snapshot_sha256":"77ffa05d879d587091933d628dde8b6687a431b7603d0a72f3b0eb71b43c25e7"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6b81a8b5fe5cd3a6cc25f321c8554800ef4ce81fe99a18c4e158fe5f79b53237"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:17.530233Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T15:00:34.596503Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"cited_work_retraction","ran_at":"2026-05-19T14:22:02.315156Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.039076Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"shingle_duplication","ran_at":"2026-05-19T13:49:41.836652Z","status":"skipped","version":"0.1.0"},{"findings_count":0,"name":"citation_quote_validity","ran_at":"2026-05-19T13:49:41.374143Z","status":"skipped","version":"0.1.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.622179Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.15529/integrity.json","findings":[],"snapshot_sha256":"1f478e0f572e050d99acf864c1daf4226687337806c1c23c782d0bda1c7ca8f7","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Process Reward Models (PRMs) provide step-level feedback for reasoning, but current PRMs usually output only a single reward score for each step. Downstream methods must therefore treat imperfect step-level reward predictions as reliable decision signals, with no indication of when these predictions should be trusted. We propose BetaPRM, a distributional PRM that predicts both a step-level success probability and the reliability of that prediction. Given step-success supervision from Monte Carlo continuations, BetaPRM learns a Beta belief that explains the observed number of successful continu","authors_text":"Chengsong Huang, Donghong Cai, Jiaxin Huang, Jinyuan Li, Langlin Huang, Shaoyang Xu, Wenxuan Zhang, Yuyi Yang","cross_cats":["cs.AI","cs.LG"],"headline":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-15T01:57:11Z","title":"Process Rewards with Learned Reliability"},"references":{"count":75,"internal_anchors":10,"resolved_work":75,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report.arXiv e-prints, pages arXiv–2502, 2025","work_id":"c10404e0-3ee3-4bc2-a540-eef0932c29b3","year":2025},{"cited_arxiv_id":"2602.01070","doi":"","is_internal_anchor":true,"ref_index":2,"title":"What If We Allocate Test-Time Compute Adaptively?","work_id":"bba84be1-f2ec-4406-8397-3397ca50235b","year":2026},{"cited_arxiv_id":"2407.21787","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","work_id":"b124064d-5a56-42ad-86f5-3cc349b86a3a","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"An augmented benchmark dataset for geometric question answering through dual parallel text encoding","work_id":"c8f950dd-cbd1-4623-9c55-46f8e51b659e","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Web-shepherd: Advancing PRMs for reinforcing web agents","work_id":"cbef34b3-6c45-4cfc-bf8b-e4e204a69eb9","year":2026}],"snapshot_sha256":"c4f5a69aea73472f4d107f940e9e114d76118724717a9c28cda8bf53336458ee"},"source":{"id":"2605.15529","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T14:44:07.430787Z","id":"0836c458-32f4-4944-8884-0a96bee9efec","model_set":{"reader":"grok-4.3"},"one_line_summary":"BetaPRM learns distributional step rewards with explicit reliability via Beta-Binomial modeling, enabling ACA that cuts token use by up to 33.57% while raising final-answer accuracy on reasoning benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search.","strongest_claim":"Experiments across four backbones and four reasoning benchmarks show that BetaPRM improves PRM-guided Best-of-N selection while preserving standard step-level error detection. Built on this signal, ACA improves the accuracy--token tradeoff over fixed-budget Best-of-16, reducing token usage by up to 33.57% while improving final-answer accuracy.","weakest_assumption":"The assumption that the Beta posterior learned from finite Monte Carlo continuations provides a calibrated and actionable measure of true prediction reliability rather than merely reflecting sampling noise or model bias in the continuation process itself (implicit in the Beta-Binomial likelihood description)."}},"verdict_id":"0836c458-32f4-4944-8884-0a96bee9efec"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:269337877b4d59123bf7ecac4c5772fd03cc15866230358a8df8b0298137fb0e","target":"record","created_at":"2026-05-20T00:01:03Z","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":"e83e96a90d1c1027b3412052968f37772dc914a154b20ad7d4f86d123f5ab6db","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-15T01:57:11Z","title_canon_sha256":"23a708fc428a0e28c2bf7fc6b9c21bee42b41a0f16031c2e27dbb82b71c0b6e5"},"schema_version":"1.0","source":{"id":"2605.15529","kind":"arxiv","version":1}},"canonical_sha256":"acc9726f7c198bb80bc4453b3bb8e1491f1854f3753e2a6961fdcdfa100e7c24","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"acc9726f7c198bb80bc4453b3bb8e1491f1854f3753e2a6961fdcdfa100e7c24","first_computed_at":"2026-05-20T00:01:03.617279Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:01:03.617279Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"A+zniaDfnnO6dMeiIYd8MuEvWpc9sicqO+TUHjc10BVgGP1uQDwEK8Dz9XwaN+PdjavdthswqoBZhmO+Bb/iDg==","signature_status":"signed_v1","signed_at":"2026-05-20T00:01:03.618220Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.15529","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:269337877b4d59123bf7ecac4c5772fd03cc15866230358a8df8b0298137fb0e","sha256:6fa2ed170304c60bd942a6fef565058c0065c4ddde03b75748d19dcd7246b0e1"],"state_sha256":"49b5ebd616c2d65f536ce727393df7253eba40b582d851913bd4de6d33e0098c"}