{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:DIZ3IBNGIF2G2ZPXB2FQ6IDDOI","short_pith_number":"pith:DIZ3IBNG","schema_version":"1.0","canonical_sha256":"1a33b405a641746d65f70e8b0f2063720362a63041f8816053880526e2ca9c47","source":{"kind":"arxiv","id":"2606.07520","version":1},"attestation_state":"computed","paper":{"title":"TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bibo Cai, Dandan Tu, Haonan Song, Qixun Zhang, Ting Liu, Wu Ning, Xiao Ding, Yirong Zeng, Yufei Liu, Yutai Hou, Yuxiang He, Yuxian Wang","submitted_at":"2026-04-19T06:02:15Z","abstract_excerpt":"Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2606.07520","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-19T06:02:15Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"6983faec42c25a46996143375fc89e86dbf50af4277b57727ce7cab6f70a3d5c","abstract_canon_sha256":"16ec8601d10f009d9248114f609965cbba09947f83f57b1702ef7ddf4c143b38"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T00:04:41.282251Z","signature_b64":"uSgel6RNQn9UTIiFJxdxXXClftxqm5EdkEPZJRomXfN0s8Mgz5Ix+HPrjGElHzOPikQo5owoSLWoz/4oXGdCDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1a33b405a641746d65f70e8b0f2063720362a63041f8816053880526e2ca9c47","last_reissued_at":"2026-06-09T00:04:41.281436Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T00:04:41.281436Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bibo Cai, Dandan Tu, Haonan Song, Qixun Zhang, Ting Liu, Wu Ning, Xiao Ding, Yirong Zeng, Yufei Liu, Yutai Hou, Yuxiang He, Yuxian Wang","submitted_at":"2026-04-19T06:02:15Z","abstract_excerpt":"Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.07520","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/2606.07520/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2606.07520","created_at":"2026-06-09T00:04:41.281553+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.07520v1","created_at":"2026-06-09T00:04:41.281553+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.07520","created_at":"2026-06-09T00:04:41.281553+00:00"},{"alias_kind":"pith_short_12","alias_value":"DIZ3IBNGIF2G","created_at":"2026-06-09T00:04:41.281553+00:00"},{"alias_kind":"pith_short_16","alias_value":"DIZ3IBNGIF2G2ZPX","created_at":"2026-06-09T00:04:41.281553+00:00"},{"alias_kind":"pith_short_8","alias_value":"DIZ3IBNG","created_at":"2026-06-09T00:04:41.281553+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI","json":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI.json","graph_json":"https://pith.science/api/pith-number/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/graph.json","events_json":"https://pith.science/api/pith-number/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/events.json","paper":"https://pith.science/paper/DIZ3IBNG"},"agent_actions":{"view_html":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI","download_json":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI.json","view_paper":"https://pith.science/paper/DIZ3IBNG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.07520&json=true","fetch_graph":"https://pith.science/api/pith-number/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/graph.json","fetch_events":"https://pith.science/api/pith-number/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/action/storage_attestation","attest_author":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/action/author_attestation","sign_citation":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/action/citation_signature","submit_replication":"https://pith.science/pith/DIZ3IBNGIF2G2ZPXB2FQ6IDDOI/action/replication_record"}},"created_at":"2026-06-09T00:04:41.281553+00:00","updated_at":"2026-06-09T00:04:41.281553+00:00"}