{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:757YGXPHWYAGOGBDYTSV2QBF46","short_pith_number":"pith:757YGXPH","canonical_record":{"source":{"id":"2604.20926","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2026-04-22T07:29:23Z","cross_cats_sorted":[],"title_canon_sha256":"45377acce903cc8fd2e3d7fee2f5cbb2b7b37e2ba8bd6c6b657800b2e7b669bf","abstract_canon_sha256":"640b77926b1e3a1cf8d6603b65e94e4ad06449e84c863c796f1527a6887a7358"},"schema_version":"1.0"},"canonical_sha256":"ff7f835de7b600671823c4e55d4025e7bf9ab52472b5524049947c919f6df730","source":{"kind":"arxiv","id":"2604.20926","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.20926","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"arxiv_version","alias_value":"2604.20926v3","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.20926","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_12","alias_value":"757YGXPHWYAG","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_16","alias_value":"757YGXPHWYAGOGBD","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_8","alias_value":"757YGXPH","created_at":"2026-06-02T02:04:17Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:757YGXPHWYAGOGBDYTSV2QBF46","target":"record","payload":{"canonical_record":{"source":{"id":"2604.20926","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2026-04-22T07:29:23Z","cross_cats_sorted":[],"title_canon_sha256":"45377acce903cc8fd2e3d7fee2f5cbb2b7b37e2ba8bd6c6b657800b2e7b669bf","abstract_canon_sha256":"640b77926b1e3a1cf8d6603b65e94e4ad06449e84c863c796f1527a6887a7358"},"schema_version":"1.0"},"canonical_sha256":"ff7f835de7b600671823c4e55d4025e7bf9ab52472b5524049947c919f6df730","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:17.923514Z","signature_b64":"TvN7AH4Hs3KAitYl937l4qQDO3ib1pTMfgVP0CbVdtlw/VeVYe0KNpZZ96Ajyu9kMguzjc/Pf5ERInH+OSuKDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ff7f835de7b600671823c4e55d4025e7bf9ab52472b5524049947c919f6df730","last_reissued_at":"2026-06-02T02:04:17.922989Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:17.922989Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.20926","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"BHpK5+Kt4zQSvorDzm4/dpEUr+3RRW3b50Yywo3VhIs5BRpqzUEcF/NS/jd9iDEDrINovUC3WbFWt6WgsCg4Bw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-07T10:48:33.799757Z"},"content_sha256":"b878244dc90a6cdb508741e1d389e620ae736d167e8b59deddcc99ef080706b9","schema_version":"1.0","event_id":"sha256:b878244dc90a6cdb508741e1d389e620ae736d167e8b59deddcc99ef080706b9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:757YGXPHWYAGOGBDYTSV2QBF46","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Learning Reasoning World Models for Parallel Code","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code.","cross_cats":[],"primary_cat":"cs.SE","authors_text":"Arjun Guha, Bhavya Kailkhura, Gautam Singh, Harshitha Menon","submitted_at":"2026-04-22T07:29:23Z","abstract_excerpt":"Large language models have shown remarkable ability in serial code generation, but they still struggle with parallel code for which training data is comparatively scarce. A common remedy is to use coding agents that interact with external tools, but tool calls can be costly and sometimes impractical, e.g., for partially written code. We propose Parallel-Code World Models (PCWMs), reasoning LLMs that aim to predict tool outcomes directly from parallel source code. To train PCWMs, we design a novel exploration and data generation pipeline that samples diverse parallel-coding problems and candida"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Fine-tuning on the resulting data yields noticeable gains, with a 7B-parameter world model improving from 64.3% to 72.8% accuracy in race-outcome prediction, while an 8B-parameter model improves in a performance profiling task from 49.3% to 58.6% accuracy. Furthermore, when open-weight models were tasked with fixing data races, world-model feedback improved their race-fixing rates relative to self-feedback by 2.7%-9.1% using our 7B-parameter world model and by 6.1%-11.1% using our 14B-parameter world model.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the synthesized hindsight reasoning traces accurately capture causal connections between source code and tool outcomes rather than superficial correlations, and that these predictions generalize beyond the sampled domains and execution environments used in data generation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Fine-tuned reasoning LLMs predict parallel code execution outcomes like data races and performance profiles, yielding accuracy gains and better race-fixing feedback than self-feedback baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"04623e47b1162cbc8c47023a59788486ae5fd26dc36eb1654f4d24399d05c77c"},"source":{"id":"2604.20926","kind":"arxiv","version":3},"verdict":{"id":"75f4fd33-4455-4d5d-bd48-0424f3efa84f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T00:14:34.828393Z","strongest_claim":"Fine-tuning on the resulting data yields noticeable gains, with a 7B-parameter world model improving from 64.3% to 72.8% accuracy in race-outcome prediction, while an 8B-parameter model improves in a performance profiling task from 49.3% to 58.6% accuracy. Furthermore, when open-weight models were tasked with fixing data races, world-model feedback improved their race-fixing rates relative to self-feedback by 2.7%-9.1% using our 7B-parameter world model and by 6.1%-11.1% using our 14B-parameter world model.","one_line_summary":"Fine-tuned reasoning LLMs predict parallel code execution outcomes like data races and performance profiles, yielding accuracy gains and better race-fixing feedback than self-feedback baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the synthesized hindsight reasoning traces accurately capture causal connections between source code and tool outcomes rather than superficial correlations, and that these predictions generalize beyond the sampled domains and execution environments used in data generation.","pith_extraction_headline":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.20926/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-21T14:43:49.996703Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-20T02:07:16.378666Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"f07982b8acc304414c84cea86382347365ecdfe500480a7111f510f971afee79"},"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":"75f4fd33-4455-4d5d-bd48-0424f3efa84f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"OHx+bQW8ypr8K+zyWuFuo5HY3KAfaq2hJ4YX7djSGMEHl2wBv4bkDj0fPDPbFDaXCbFbLUUsxYIH6gmJFSbXDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-07T10:48:33.800608Z"},"content_sha256":"6826c878bd287eea67bc8756440a6556262430edab2d0ad089c0403625659ec9","schema_version":"1.0","event_id":"sha256:6826c878bd287eea67bc8756440a6556262430edab2d0ad089c0403625659ec9"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/757YGXPHWYAGOGBDYTSV2QBF46/bundle.json","state_url":"https://pith.science/pith/757YGXPHWYAGOGBDYTSV2QBF46/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/757YGXPHWYAGOGBDYTSV2QBF46/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":"2026-06-07T10:48:33Z","links":{"resolver":"https://pith.science/pith/757YGXPHWYAGOGBDYTSV2QBF46","bundle":"https://pith.science/pith/757YGXPHWYAGOGBDYTSV2QBF46/bundle.json","state":"https://pith.science/pith/757YGXPHWYAGOGBDYTSV2QBF46/state.json","well_known_bundle":"https://pith.science/.well-known/pith/757YGXPHWYAGOGBDYTSV2QBF46/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:757YGXPHWYAGOGBDYTSV2QBF46","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":"640b77926b1e3a1cf8d6603b65e94e4ad06449e84c863c796f1527a6887a7358","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2026-04-22T07:29:23Z","title_canon_sha256":"45377acce903cc8fd2e3d7fee2f5cbb2b7b37e2ba8bd6c6b657800b2e7b669bf"},"schema_version":"1.0","source":{"id":"2604.20926","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.20926","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"arxiv_version","alias_value":"2604.20926v3","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.20926","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_12","alias_value":"757YGXPHWYAG","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_16","alias_value":"757YGXPHWYAGOGBD","created_at":"2026-06-02T02:04:17Z"},{"alias_kind":"pith_short_8","alias_value":"757YGXPH","created_at":"2026-06-02T02:04:17Z"}],"graph_snapshots":[{"event_id":"sha256:6826c878bd287eea67bc8756440a6556262430edab2d0ad089c0403625659ec9","target":"graph","created_at":"2026-06-02T02:04:17Z","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":"Fine-tuning on the resulting data yields noticeable gains, with a 7B-parameter world model improving from 64.3% to 72.8% accuracy in race-outcome prediction, while an 8B-parameter model improves in a performance profiling task from 49.3% to 58.6% accuracy. Furthermore, when open-weight models were tasked with fixing data races, world-model feedback improved their race-fixing rates relative to self-feedback by 2.7%-9.1% using our 7B-parameter world model and by 6.1%-11.1% using our 14B-parameter world model."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the synthesized hindsight reasoning traces accurately capture causal connections between source code and tool outcomes rather than superficial correlations, and that these predictions generalize beyond the sampled domains and execution environments used in data generation."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Fine-tuned reasoning LLMs predict parallel code execution outcomes like data races and performance profiles, yielding accuracy gains and better race-fixing feedback than self-feedback baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code."}],"snapshot_sha256":"04623e47b1162cbc8c47023a59788486ae5fd26dc36eb1654f4d24399d05c77c"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-21T14:43:49.996703Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T02:07:16.378666Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2604.20926/integrity.json","findings":[],"snapshot_sha256":"f07982b8acc304414c84cea86382347365ecdfe500480a7111f510f971afee79","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models have shown remarkable ability in serial code generation, but they still struggle with parallel code for which training data is comparatively scarce. A common remedy is to use coding agents that interact with external tools, but tool calls can be costly and sometimes impractical, e.g., for partially written code. We propose Parallel-Code World Models (PCWMs), reasoning LLMs that aim to predict tool outcomes directly from parallel source code. To train PCWMs, we design a novel exploration and data generation pipeline that samples diverse parallel-coding problems and candida","authors_text":"Arjun Guha, Bhavya Kailkhura, Gautam Singh, Harshitha Menon","cross_cats":[],"headline":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2026-04-22T07:29:23Z","title":"Learning Reasoning World Models for Parallel Code"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.20926","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-10T00:14:34.828393Z","id":"75f4fd33-4455-4d5d-bd48-0424f3efa84f","model_set":{"reader":"grok-4.3"},"one_line_summary":"Fine-tuned reasoning LLMs predict parallel code execution outcomes like data races and performance profiles, yielding accuracy gains and better race-fixing feedback than self-feedback baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Reasoning language models can be trained to predict parallel code outcomes such as data races and performance profiles directly from source code.","strongest_claim":"Fine-tuning on the resulting data yields noticeable gains, with a 7B-parameter world model improving from 64.3% to 72.8% accuracy in race-outcome prediction, while an 8B-parameter model improves in a performance profiling task from 49.3% to 58.6% accuracy. Furthermore, when open-weight models were tasked with fixing data races, world-model feedback improved their race-fixing rates relative to self-feedback by 2.7%-9.1% using our 7B-parameter world model and by 6.1%-11.1% using our 14B-parameter world model.","weakest_assumption":"That the synthesized hindsight reasoning traces accurately capture causal connections between source code and tool outcomes rather than superficial correlations, and that these predictions generalize beyond the sampled domains and execution environments used in data generation."}},"verdict_id":"75f4fd33-4455-4d5d-bd48-0424f3efa84f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b878244dc90a6cdb508741e1d389e620ae736d167e8b59deddcc99ef080706b9","target":"record","created_at":"2026-06-02T02:04:17Z","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":"640b77926b1e3a1cf8d6603b65e94e4ad06449e84c863c796f1527a6887a7358","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2026-04-22T07:29:23Z","title_canon_sha256":"45377acce903cc8fd2e3d7fee2f5cbb2b7b37e2ba8bd6c6b657800b2e7b669bf"},"schema_version":"1.0","source":{"id":"2604.20926","kind":"arxiv","version":3}},"canonical_sha256":"ff7f835de7b600671823c4e55d4025e7bf9ab52472b5524049947c919f6df730","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ff7f835de7b600671823c4e55d4025e7bf9ab52472b5524049947c919f6df730","first_computed_at":"2026-06-02T02:04:17.922989Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-02T02:04:17.922989Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"TvN7AH4Hs3KAitYl937l4qQDO3ib1pTMfgVP0CbVdtlw/VeVYe0KNpZZ96Ajyu9kMguzjc/Pf5ERInH+OSuKDA==","signature_status":"signed_v1","signed_at":"2026-06-02T02:04:17.923514Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.20926","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b878244dc90a6cdb508741e1d389e620ae736d167e8b59deddcc99ef080706b9","sha256:6826c878bd287eea67bc8756440a6556262430edab2d0ad089c0403625659ec9"],"state_sha256":"627bdc3ed772bd8f938026fd517279deed7872aa81e1731cf305b497b412f709"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"elJWtrTgzd4kg8cqHY3wft5MEvgWMsga5n+ODHOza0R64gSjVhpMz4vuE7E23Bm4HELZDM+qc4SKvWzNnIfICg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-07T10:48:33.804242Z","bundle_sha256":"e854e7484858f8582066cc2856ebaf4e34a16b3c848a7218e450f113dcd877cf"}}