{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:HJKVI4V7UFKRPPTEV6GNMYAZK5","short_pith_number":"pith:HJKVI4V7","canonical_record":{"source":{"id":"2602.06462","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-06T07:47:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"36ac30f0e816fd51be5f08dc8508828533c774c21170651e941191c1db7db177","abstract_canon_sha256":"c58177542f6861c6af0feda73a415a68217556e9e02e8a7b1736588eb26f3914"},"schema_version":"1.0"},"canonical_sha256":"3a555472bfa15517be64af8cd66019574ac11cbdf4ecc44b53a50e93d6d71480","source":{"kind":"arxiv","id":"2602.06462","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.06462","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"arxiv_version","alias_value":"2602.06462v4","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.06462","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_12","alias_value":"HJKVI4V7UFKR","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_16","alias_value":"HJKVI4V7UFKRPPTE","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_8","alias_value":"HJKVI4V7","created_at":"2026-05-20T01:05:08Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:HJKVI4V7UFKRPPTEV6GNMYAZK5","target":"record","payload":{"canonical_record":{"source":{"id":"2602.06462","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-06T07:47:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"36ac30f0e816fd51be5f08dc8508828533c774c21170651e941191c1db7db177","abstract_canon_sha256":"c58177542f6861c6af0feda73a415a68217556e9e02e8a7b1736588eb26f3914"},"schema_version":"1.0"},"canonical_sha256":"3a555472bfa15517be64af8cd66019574ac11cbdf4ecc44b53a50e93d6d71480","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:08.345875Z","signature_b64":"fepWEfMiop3SgWEqt15Olz1aKJO3hc7HspBiBTil/OP5UfXk06KOQgiNAIEYg+vHVw2DP+KlKOr8dNlVGjdCCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a555472bfa15517be64af8cd66019574ac11cbdf4ecc44b53a50e93d6d71480","last_reissued_at":"2026-05-20T01:05:08.344969Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:08.344969Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.06462","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T01:05:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6jVMXdsPYFtkIv0tjjPZMp628pV0BQyLd8DxxUZfs6HaPzVV6qfckHFu1nw5JRRVKbMNpRYTJbNJEykM6To+Aw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T12:46:03.896401Z"},"content_sha256":"5688afe26f6aea48e8f72634c9dd098446fe1b6d789b2789f063d6c27ab8f96e","schema_version":"1.0","event_id":"sha256:5688afe26f6aea48e8f72634c9dd098446fe1b6d789b2789f063d6c27ab8f96e"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:HJKVI4V7UFKRPPTEV6GNMYAZK5","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Diffusion-State Policy Optimization for Masked Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Daisuke Oba, Hiroki Furuta, Naoaki Okazaki","submitted_at":"2026-02-06T07:47:22Z","abstract_excerpt":"Masked diffusion language models generate text through iterative masked-token filling, but terminal-only rewards on final completions provide coarse credit assignment for the intermediate filling decisions that shape the generation process. We propose Diffusion-State Policy Optimization (DiSPO), a plug-in credit-assignment layer that directly optimizes intermediate filling decisions. At selected intermediate masked states, DiSPO branches by resampling the currently masked positions from rollout-cached logits, scores the resulting completions, and updates only the newly filled tokens, requiring"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DiSPO consistently improves terminal-feedback baselines, including diffu-GRPO and SPG, on math and planning benchmarks under matched rollout compute and optimizer steps.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That resampling completions from rollout-cached logits at selected intermediate masked states yields unbiased or sufficiently low-variance credit signals for the newly filled tokens without introducing systematic bias from the branching process.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DiSPO is a plug-in credit-assignment method for masked diffusion LMs that optimizes intermediate filling decisions via branched completions from rollout-cached logits.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"37c4f7d682aca412ebd4ecd92663bcf88321866378701aeb0f74508e59301598"},"source":{"id":"2602.06462","kind":"arxiv","version":4},"verdict":{"id":"496b9848-ea59-44b0-a886-50d9cea36c5d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T07:14:48.723580Z","strongest_claim":"DiSPO consistently improves terminal-feedback baselines, including diffu-GRPO and SPG, on math and planning benchmarks under matched rollout compute and optimizer steps.","one_line_summary":"DiSPO is a plug-in credit-assignment method for masked diffusion LMs that optimizes intermediate filling decisions via branched completions from rollout-cached logits.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That resampling completions from rollout-cached logits at selected intermediate masked states yields unbiased or sufficiently low-variance credit signals for the newly filled tokens without introducing systematic bias from the branching process.","pith_extraction_headline":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.06462/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":"496b9848-ea59-44b0-a886-50d9cea36c5d"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T01:05:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"+rwGb0KOUGGDKqzS1a3iCNRjIoYgDWDNwibAFF6MdoaIMGa5fCnyUomWp+mupof9/HR7NBtcpRkLzZDqVmpDBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-09T12:46:03.896869Z"},"content_sha256":"42d67227fc4e11413b8e160bda57fb17a088f02b5537275aad5ee277b365e693","schema_version":"1.0","event_id":"sha256:42d67227fc4e11413b8e160bda57fb17a088f02b5537275aad5ee277b365e693"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/bundle.json","state_url":"https://pith.science/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/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":"2026-06-09T12:46:03Z","links":{"resolver":"https://pith.science/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5","bundle":"https://pith.science/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/bundle.json","state":"https://pith.science/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HJKVI4V7UFKRPPTEV6GNMYAZK5/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:HJKVI4V7UFKRPPTEV6GNMYAZK5","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":"c58177542f6861c6af0feda73a415a68217556e9e02e8a7b1736588eb26f3914","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-06T07:47:22Z","title_canon_sha256":"36ac30f0e816fd51be5f08dc8508828533c774c21170651e941191c1db7db177"},"schema_version":"1.0","source":{"id":"2602.06462","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.06462","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"arxiv_version","alias_value":"2602.06462v4","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.06462","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_12","alias_value":"HJKVI4V7UFKR","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_16","alias_value":"HJKVI4V7UFKRPPTE","created_at":"2026-05-20T01:05:08Z"},{"alias_kind":"pith_short_8","alias_value":"HJKVI4V7","created_at":"2026-05-20T01:05:08Z"}],"graph_snapshots":[{"event_id":"sha256:42d67227fc4e11413b8e160bda57fb17a088f02b5537275aad5ee277b365e693","target":"graph","created_at":"2026-05-20T01:05:08Z","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":"DiSPO consistently improves terminal-feedback baselines, including diffu-GRPO and SPG, on math and planning benchmarks under matched rollout compute and optimizer steps."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That resampling completions from rollout-cached logits at selected intermediate masked states yields unbiased or sufficiently low-variance credit signals for the newly filled tokens without introducing systematic bias from the branching process."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"DiSPO is a plug-in credit-assignment method for masked diffusion LMs that optimizes intermediate filling decisions via branched completions from rollout-cached logits."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits."}],"snapshot_sha256":"37c4f7d682aca412ebd4ecd92663bcf88321866378701aeb0f74508e59301598"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2602.06462/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Masked diffusion language models generate text through iterative masked-token filling, but terminal-only rewards on final completions provide coarse credit assignment for the intermediate filling decisions that shape the generation process. We propose Diffusion-State Policy Optimization (DiSPO), a plug-in credit-assignment layer that directly optimizes intermediate filling decisions. At selected intermediate masked states, DiSPO branches by resampling the currently masked positions from rollout-cached logits, scores the resulting completions, and updates only the newly filled tokens, requiring","authors_text":"Daisuke Oba, Hiroki Furuta, Naoaki Okazaki","cross_cats":["cs.LG"],"headline":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-06T07:47:22Z","title":"Diffusion-State Policy Optimization for Masked Diffusion Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.06462","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-16T07:14:48.723580Z","id":"496b9848-ea59-44b0-a886-50d9cea36c5d","model_set":{"reader":"grok-4.3"},"one_line_summary":"DiSPO is a plug-in credit-assignment method for masked diffusion LMs that optimizes intermediate filling decisions via branched completions from rollout-cached logits.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"DiSPO assigns credit to intermediate token-filling decisions in masked diffusion language models by branching from cached logits.","strongest_claim":"DiSPO consistently improves terminal-feedback baselines, including diffu-GRPO and SPG, on math and planning benchmarks under matched rollout compute and optimizer steps.","weakest_assumption":"That resampling completions from rollout-cached logits at selected intermediate masked states yields unbiased or sufficiently low-variance credit signals for the newly filled tokens without introducing systematic bias from the branching process."}},"verdict_id":"496b9848-ea59-44b0-a886-50d9cea36c5d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:5688afe26f6aea48e8f72634c9dd098446fe1b6d789b2789f063d6c27ab8f96e","target":"record","created_at":"2026-05-20T01:05:08Z","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":"c58177542f6861c6af0feda73a415a68217556e9e02e8a7b1736588eb26f3914","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-06T07:47:22Z","title_canon_sha256":"36ac30f0e816fd51be5f08dc8508828533c774c21170651e941191c1db7db177"},"schema_version":"1.0","source":{"id":"2602.06462","kind":"arxiv","version":4}},"canonical_sha256":"3a555472bfa15517be64af8cd66019574ac11cbdf4ecc44b53a50e93d6d71480","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"3a555472bfa15517be64af8cd66019574ac11cbdf4ecc44b53a50e93d6d71480","first_computed_at":"2026-05-20T01:05:08.344969Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T01:05:08.344969Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"fepWEfMiop3SgWEqt15Olz1aKJO3hc7HspBiBTil/OP5UfXk06KOQgiNAIEYg+vHVw2DP+KlKOr8dNlVGjdCCA==","signature_status":"signed_v1","signed_at":"2026-05-20T01:05:08.345875Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.06462","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:5688afe26f6aea48e8f72634c9dd098446fe1b6d789b2789f063d6c27ab8f96e","sha256:42d67227fc4e11413b8e160bda57fb17a088f02b5537275aad5ee277b365e693"],"state_sha256":"9ff9224976f5129b2bf720dbb9e29340a3d8a9088955292fa13f22fd71050115"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"jPnNVZb1L9Y8uQhMJdoGfuubBG8qwklaspjD/7w8DZ7yIa7Px1uVZWspF7L7BmsBMA7Xr+G81QQwcz+RAF8ZBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-09T12:46:03.899232Z","bundle_sha256":"74a63b462e46ab5f0473374d31152b4d0f9b86b5a0ee6e6a6695522fa80a170e"}}