{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:UEGF26OPBHXWQYWCUZ3E6DY4QH","short_pith_number":"pith:UEGF26OP","canonical_record":{"source":{"id":"2605.07210","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-08T03:57:51Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"9b5b02bb7d1878c4d81457911f6b24b1648068755c90ecfc9fc996f0b714db11","abstract_canon_sha256":"9784ac58fbf37def983e49f3e3716a2a4cca753cbd7c340cb41985c0569fe5cc"},"schema_version":"1.0"},"canonical_sha256":"a10c5d79cf09ef6862c2a6764f0f1c81f906cafada29ce3f2b513919273feb84","source":{"kind":"arxiv","id":"2605.07210","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.07210","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"arxiv_version","alias_value":"2605.07210v2","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.07210","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_12","alias_value":"UEGF26OPBHXW","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_16","alias_value":"UEGF26OPBHXWQYWC","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_8","alias_value":"UEGF26OP","created_at":"2026-05-29T01:04:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:UEGF26OPBHXWQYWCUZ3E6DY4QH","target":"record","payload":{"canonical_record":{"source":{"id":"2605.07210","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-08T03:57:51Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"9b5b02bb7d1878c4d81457911f6b24b1648068755c90ecfc9fc996f0b714db11","abstract_canon_sha256":"9784ac58fbf37def983e49f3e3716a2a4cca753cbd7c340cb41985c0569fe5cc"},"schema_version":"1.0"},"canonical_sha256":"a10c5d79cf09ef6862c2a6764f0f1c81f906cafada29ce3f2b513919273feb84","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:04:37.620846Z","signature_b64":"uI3gEq/mPTDJZE3uWqDbObkGjiwThPYsjaai4VWxI2XIOOD57rXDSv5HtUcR9lx236VNt/G+5SULyEVRn3uPDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a10c5d79cf09ef6862c2a6764f0f1c81f906cafada29ce3f2b513919273feb84","last_reissued_at":"2026-05-29T01:04:37.620235Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:04:37.620235Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.07210","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-29T01:04:37Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"5/43NufgnPXZAk5xXJP31tnpoZtTwYAjoQk4sG14bYz+Ji4r9eGDlVh7zpizrv4Cv/oXwObd9uWh8GBStOc1AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T19:45:40.381296Z"},"content_sha256":"b6812207928c4cbba0a565bf6f7521bc9352224a05b42ba325dffb505810c28a","schema_version":"1.0","event_id":"sha256:b6812207928c4cbba0a565bf6f7521bc9352224a05b42ba325dffb505810c28a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:UEGF26OPBHXWQYWCUZ3E6DY4QH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"DiffRetriever: Parallel Representative Tokens for Retrieval with Diffusion Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning.","cross_cats":["cs.CL"],"primary_cat":"cs.IR","authors_text":"Bevan Koopman, Guido Zuccon, Shengyao Zhuang, Shuai Wang, Yu Yin","submitted_at":"2026-05-08T03:57:51Z","abstract_excerpt":"This paper shows how diffusion language models (DLMs) can be used as effective and efficient retrievers. Existing DLM-based retrievers (e.g., DiffEmbed) follow BERT-style encoding, representing each query or passage as a single mean-pooled vector. This ignores how DLMs are trained to generate responses through masked-position prediction under bidirectional attention, a capability that can provide stronger retrieval signals. We propose DiffRetriever, which uses the DLM's native masked-position prediction directly for retrieval. For each query or passage, DiffRetriever appends one or more masked"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"After supervised fine-tuning, DiffRetriever on Dream is the strongest BEIR-7 retriever in our comparison, ahead of PromptReps, the encoder-style DiffEmbed baseline on the same diffusion backbones, and the contrastively fine-tuned single-vector RepLLaMA.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed gains are attributable to the parallel multi-token mechanism enabled by diffusion rather than to differences in backbone capacity, fine-tuning procedure, or unstated implementation details; the abstract does not detail controls for these factors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DiffRetriever generates multiple representative tokens in parallel using diffusion language models, yielding consistent retrieval gains over single-token baselines and autoregressive multi-token variants on BEIR benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d249b908ba2162c7bcb8e62041af271af1857f520646cc4aee49ce1727ce15d1"},"source":{"id":"2605.07210","kind":"arxiv","version":2},"verdict":{"id":"71b92980-6fdc-4444-8956-a0a164138cf3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T01:38:02.900956Z","strongest_claim":"After supervised fine-tuning, DiffRetriever on Dream is the strongest BEIR-7 retriever in our comparison, ahead of PromptReps, the encoder-style DiffEmbed baseline on the same diffusion backbones, and the contrastively fine-tuned single-vector RepLLaMA.","one_line_summary":"DiffRetriever generates multiple representative tokens in parallel using diffusion language models, yielding consistent retrieval gains over single-token baselines and autoregressive multi-token variants on BEIR benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed gains are attributable to the parallel multi-token mechanism enabled by diffusion rather than to differences in backbone capacity, fine-tuning procedure, or unstated implementation details; the abstract does not detail controls for these factors.","pith_extraction_headline":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.07210/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T11:22:03.313302Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T06:34:52.842895Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T17:01:19.548580Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T12:00:51.385009Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"fa2583b243e98056964583be1ca9022b8bb9dae71c4ef3d65ac00a231606d8ab"},"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":"71b92980-6fdc-4444-8956-a0a164138cf3"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-29T01:04:37Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"8RmGLPWXniwqJwdRqm7xbYZhGSD047DHG+HHT71UCW/9TzFwI6gg03JKOkSWJOvQae4QvvKNF3nEpHG9pO30DQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T19:45:40.381824Z"},"content_sha256":"a202e55e151d4a7379b3d8cde79c6059a352300eaa5e6f040151e177dd959d23","schema_version":"1.0","event_id":"sha256:a202e55e151d4a7379b3d8cde79c6059a352300eaa5e6f040151e177dd959d23"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/bundle.json","state_url":"https://pith.science/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/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":"2026-06-08T19:45:40Z","links":{"resolver":"https://pith.science/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH","bundle":"https://pith.science/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/bundle.json","state":"https://pith.science/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/UEGF26OPBHXWQYWCUZ3E6DY4QH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:UEGF26OPBHXWQYWCUZ3E6DY4QH","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":"9784ac58fbf37def983e49f3e3716a2a4cca753cbd7c340cb41985c0569fe5cc","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-08T03:57:51Z","title_canon_sha256":"9b5b02bb7d1878c4d81457911f6b24b1648068755c90ecfc9fc996f0b714db11"},"schema_version":"1.0","source":{"id":"2605.07210","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.07210","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"arxiv_version","alias_value":"2605.07210v2","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.07210","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_12","alias_value":"UEGF26OPBHXW","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_16","alias_value":"UEGF26OPBHXWQYWC","created_at":"2026-05-29T01:04:37Z"},{"alias_kind":"pith_short_8","alias_value":"UEGF26OP","created_at":"2026-05-29T01:04:37Z"}],"graph_snapshots":[{"event_id":"sha256:a202e55e151d4a7379b3d8cde79c6059a352300eaa5e6f040151e177dd959d23","target":"graph","created_at":"2026-05-29T01:04:37Z","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":"After supervised fine-tuning, DiffRetriever on Dream is the strongest BEIR-7 retriever in our comparison, ahead of PromptReps, the encoder-style DiffEmbed baseline on the same diffusion backbones, and the contrastively fine-tuned single-vector RepLLaMA."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the observed gains are attributable to the parallel multi-token mechanism enabled by diffusion rather than to differences in backbone capacity, fine-tuning procedure, or unstated implementation details; the abstract does not detail controls for these factors."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"DiffRetriever generates multiple representative tokens in parallel using diffusion language models, yielding consistent retrieval gains over single-token baselines and autoregressive multi-token variants on BEIR benchmarks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning."}],"snapshot_sha256":"d249b908ba2162c7bcb8e62041af271af1857f520646cc4aee49ce1727ce15d1"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T11:22:03.313302Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T06:34:52.842895Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T17:01:19.548580Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T12:00:51.385009Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.07210/integrity.json","findings":[],"snapshot_sha256":"fa2583b243e98056964583be1ca9022b8bb9dae71c4ef3d65ac00a231606d8ab","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"This paper shows how diffusion language models (DLMs) can be used as effective and efficient retrievers. Existing DLM-based retrievers (e.g., DiffEmbed) follow BERT-style encoding, representing each query or passage as a single mean-pooled vector. This ignores how DLMs are trained to generate responses through masked-position prediction under bidirectional attention, a capability that can provide stronger retrieval signals. We propose DiffRetriever, which uses the DLM's native masked-position prediction directly for retrieval. For each query or passage, DiffRetriever appends one or more masked","authors_text":"Bevan Koopman, Guido Zuccon, Shengyao Zhuang, Shuai Wang, Yu Yin","cross_cats":["cs.CL"],"headline":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-08T03:57:51Z","title":"DiffRetriever: Parallel Representative Tokens for Retrieval with Diffusion Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.07210","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-11T01:38:02.900956Z","id":"71b92980-6fdc-4444-8956-a0a164138cf3","model_set":{"reader":"grok-4.3"},"one_line_summary":"DiffRetriever generates multiple representative tokens in parallel using diffusion language models, yielding consistent retrieval gains over single-token baselines and autoregressive multi-token variants on BEIR benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Diffusion language models generate multiple representative tokens for retrieval in a single parallel pass, improving over single-token and autoregressive methods on BEIR-7 after fine-tuning.","strongest_claim":"After supervised fine-tuning, DiffRetriever on Dream is the strongest BEIR-7 retriever in our comparison, ahead of PromptReps, the encoder-style DiffEmbed baseline on the same diffusion backbones, and the contrastively fine-tuned single-vector RepLLaMA.","weakest_assumption":"That the observed gains are attributable to the parallel multi-token mechanism enabled by diffusion rather than to differences in backbone capacity, fine-tuning procedure, or unstated implementation details; the abstract does not detail controls for these factors."}},"verdict_id":"71b92980-6fdc-4444-8956-a0a164138cf3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b6812207928c4cbba0a565bf6f7521bc9352224a05b42ba325dffb505810c28a","target":"record","created_at":"2026-05-29T01:04:37Z","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":"9784ac58fbf37def983e49f3e3716a2a4cca753cbd7c340cb41985c0569fe5cc","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-08T03:57:51Z","title_canon_sha256":"9b5b02bb7d1878c4d81457911f6b24b1648068755c90ecfc9fc996f0b714db11"},"schema_version":"1.0","source":{"id":"2605.07210","kind":"arxiv","version":2}},"canonical_sha256":"a10c5d79cf09ef6862c2a6764f0f1c81f906cafada29ce3f2b513919273feb84","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a10c5d79cf09ef6862c2a6764f0f1c81f906cafada29ce3f2b513919273feb84","first_computed_at":"2026-05-29T01:04:37.620235Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-29T01:04:37.620235Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"uI3gEq/mPTDJZE3uWqDbObkGjiwThPYsjaai4VWxI2XIOOD57rXDSv5HtUcR9lx236VNt/G+5SULyEVRn3uPDQ==","signature_status":"signed_v1","signed_at":"2026-05-29T01:04:37.620846Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.07210","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b6812207928c4cbba0a565bf6f7521bc9352224a05b42ba325dffb505810c28a","sha256:a202e55e151d4a7379b3d8cde79c6059a352300eaa5e6f040151e177dd959d23"],"state_sha256":"cef2059ca50d7f1ebb1d49ad7b53e1005d99f392da9c8c0f82283ba801f3fd0e"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"p2vdH3Kd+X7v2d3cAJxPeLrQpkgDyrZVeVVC1pr9FpGfR/Rfw975rW8ngyUioXBk0/jrBjheHQ+eBP+pa28YDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-08T19:45:40.385131Z","bundle_sha256":"817bc5b76e50225fed49b49702f6fab0d013f2557526adef06caf76b68816831"}}