{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VTEXE334DGF3QC6EIU5TXOHBJE","short_pith_number":"pith:VTEXE334","schema_version":"1.0","canonical_sha256":"acc9726f7c198bb80bc4453b3bb8e1491f1854f3753e2a6961fdcdfa100e7c24","source":{"kind":"arxiv","id":"2605.15529","version":1},"attestation_state":"computed","paper":{"title":"Process Rewards with Learned Reliability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengsong Huang, Donghong Cai, Jiaxin Huang, Jinyuan Li, Langlin Huang, Shaoyang Xu, Wenxuan Zhang, Yuyi Yang","submitted_at":"2026-05-15T01:57:11Z","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"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15529","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-15T01:57:11Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"23a708fc428a0e28c2bf7fc6b9c21bee42b41a0f16031c2e27dbb82b71c0b6e5","abstract_canon_sha256":"e83e96a90d1c1027b3412052968f37772dc914a154b20ad7d4f86d123f5ab6db"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:01:03.618220Z","signature_b64":"A+zniaDfnnO6dMeiIYd8MuEvWpc9sicqO+TUHjc10BVgGP1uQDwEK8Dz9XwaN+PdjavdthswqoBZhmO+Bb/iDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"acc9726f7c198bb80bc4453b3bb8e1491f1854f3753e2a6961fdcdfa100e7c24","last_reissued_at":"2026-05-20T00:01:03.617279Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:01:03.617279Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Process Rewards with Learned Reliability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengsong Huang, Donghong Cai, Jiaxin Huang, Jinyuan Li, Langlin Huang, Shaoyang Xu, Wenxuan Zhang, Yuyi Yang","submitted_at":"2026-05-15T01:57:11Z","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"},"claims":{"count":4,"items":[{"kind":"strongest_claim","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.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","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).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","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.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"77ffa05d879d587091933d628dde8b6687a431b7603d0a72f3b0eb71b43c25e7"},"source":{"id":"2605.15529","kind":"arxiv","version":1},"verdict":{"id":"0836c458-32f4-4944-8884-0a96bee9efec","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T14:44:07.430787Z","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.","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","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).","pith_extraction_headline":"A process reward model learns both step success probability and the reliability of that probability to guide more efficient reasoning search."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15529/integrity.json","findings":[],"available":true,"detectors_run":[{"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","findings_count":0}],"snapshot_sha256":"1f478e0f572e050d99acf864c1daf4226687337806c1c23c782d0bda1c7ca8f7"},"references":{"count":75,"sample":[{"doi":"","year":2025,"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","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"What If We Allocate Test-Time Compute Adaptively?","work_id":"bba84be1-f2ec-4406-8397-3397ca50235b","ref_index":2,"cited_arxiv_id":"2602.01070","is_internal_anchor":true},{"doi":"","year":2024,"title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","work_id":"b124064d-5a56-42ad-86f5-3cc349b86a3a","ref_index":3,"cited_arxiv_id":"2407.21787","is_internal_anchor":true},{"doi":"","year":2022,"title":"An augmented benchmark dataset for geometric question answering through dual parallel text encoding","work_id":"c8f950dd-cbd1-4623-9c55-46f8e51b659e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Web-shepherd: Advancing PRMs for reinforcing web agents","work_id":"cbef34b3-6c45-4cfc-bf8b-e4e204a69eb9","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":75,"snapshot_sha256":"c4f5a69aea73472f4d107f940e9e114d76118724717a9c28cda8bf53336458ee","internal_anchors":10},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6b81a8b5fe5cd3a6cc25f321c8554800ef4ce81fe99a18c4e158fe5f79b53237"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.15529","created_at":"2026-05-20T00:01:03.617453+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15529v1","created_at":"2026-05-20T00:01:03.617453+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15529","created_at":"2026-05-20T00:01:03.617453+00:00"},{"alias_kind":"pith_short_12","alias_value":"VTEXE334DGF3","created_at":"2026-05-20T00:01:03.617453+00:00"},{"alias_kind":"pith_short_16","alias_value":"VTEXE334DGF3QC6E","created_at":"2026-05-20T00:01:03.617453+00:00"},{"alias_kind":"pith_short_8","alias_value":"VTEXE334","created_at":"2026-05-20T00:01:03.617453+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE","json":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE.json","graph_json":"https://pith.science/api/pith-number/VTEXE334DGF3QC6EIU5TXOHBJE/graph.json","events_json":"https://pith.science/api/pith-number/VTEXE334DGF3QC6EIU5TXOHBJE/events.json","paper":"https://pith.science/paper/VTEXE334"},"agent_actions":{"view_html":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE","download_json":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE.json","view_paper":"https://pith.science/paper/VTEXE334","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15529&json=true","fetch_graph":"https://pith.science/api/pith-number/VTEXE334DGF3QC6EIU5TXOHBJE/graph.json","fetch_events":"https://pith.science/api/pith-number/VTEXE334DGF3QC6EIU5TXOHBJE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE/action/storage_attestation","attest_author":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE/action/author_attestation","sign_citation":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE/action/citation_signature","submit_replication":"https://pith.science/pith/VTEXE334DGF3QC6EIU5TXOHBJE/action/replication_record"}},"created_at":"2026-05-20T00:01:03.617453+00:00","updated_at":"2026-05-20T00:01:03.617453+00:00"}