{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VS7FSX64WHQIURZFLXCWM6ZMSF","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":"323ee04d9b2ad9d88a7635b939cf925e81804fdeac22020879fc7707e7109867","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T21:31:19Z","title_canon_sha256":"45db951a6c79865d598c4d8bd1179df722501ed8c78f2d4a3dec3ec011210a1c"},"schema_version":"1.0","source":{"id":"2605.15436","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.15436","created_at":"2026-05-20T00:00:58Z"},{"alias_kind":"arxiv_version","alias_value":"2605.15436v1","created_at":"2026-05-20T00:00:58Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15436","created_at":"2026-05-20T00:00:58Z"},{"alias_kind":"pith_short_12","alias_value":"VS7FSX64WHQI","created_at":"2026-05-20T00:00:58Z"},{"alias_kind":"pith_short_16","alias_value":"VS7FSX64WHQIURZF","created_at":"2026-05-20T00:00:58Z"},{"alias_kind":"pith_short_8","alias_value":"VS7FSX64","created_at":"2026-05-20T00:00:58Z"}],"graph_snapshots":[{"event_id":"sha256:dfa1d7a17ed1306df020a38991b7dbec6944faa7904bf3e1f316c56988bf84d9","target":"graph","created_at":"2026-05-20T00:00:58Z","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":"Our analysis of 144 task-model combinations demonstrates that mathematical reasoning consistently produces the highest attention entropy across all architectures, while decoder models exhibit significantly higher sparsity patterns compared to encoder models."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The twelve cognitive task categories and the chosen measurement definitions (final activation values, attention entropy, sparsity) are assumed to capture meaningful and comparable computational differences without substantial confounding from task formulation or model-specific tokenization effects."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Analysis of 144 task-model pairs finds mathematical reasoning produces the highest attention entropy in all architectures while decoder models show significantly higher sparsity than encoders."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Mathematical reasoning produces the highest attention entropy across language model architectures."}],"snapshot_sha256":"cb753b625d5120054f5a9be11938aa2feab358259e69e383962b7e0ca8d75b9c"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"cited_work_retraction","ran_at":"2026-05-19T15:54:27.873474Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"citation_quote_validity","ran_at":"2026-05-19T15:50:28.623448Z","status":"completed","version":"0.1.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T15:10:45.213638Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:17.674760Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.125219Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.690326Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.15436/integrity.json","findings":[],"snapshot_sha256":"09d13f069fe283c64d2763275445386754e874eb0bdf91e0a9ab2df981505d11","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"This paper presents a comprehensive analysis of neural activation patterns across six distinct large language model (LLM) architectures, examining their performance on twelve cognitive task categories. Through systematic measurement of final activation values, attention entropy, and sparsity patterns, we reveal fundamental differences in how encoder and decoder architectures process diverse cognitive tasks. Our analysis of 144 task-model combinations demonstrates that mathematical reasoning consistently produces the highest attention entropy across all architectures, while decoder models exhib","authors_text":"Faezeh Ghaderi, Mahdi Naser-Moghadasi","cross_cats":["cs.LG"],"headline":"Mathematical reasoning produces the highest attention entropy across language model architectures.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T21:31:19Z","title":"Neural Activation Patterns Across Language Model Architectures: A Comprehensive Analysis of Cognitive Task Performance"},"references":{"count":50,"internal_anchors":10,"resolved_work":50,"sample":[{"cited_arxiv_id":"2307.09288","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","year":2023},{"cited_arxiv_id":"2310.06825","doi":"","is_internal_anchor":true,"ref_index":2,"title":"A. Q. Jiang et al., ”Mistral 7B,” arXiv preprint arXiv:2310.06825, 2023","work_id":"eb5e1305-ad11-4875-ad8d-ad8b8f697599","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"J. Devlin, M. Chang, K. Lee, and K. Toutanova, ”BERT: Pre-training of deep bidirectional transformers for language understanding,” in Pro- ceedings of NAACL-HLT, 2019, pp. 4171-4186","work_id":"35880d3c-8917-4682-ab49-f3c2d7a4c8b4","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Radford et al., ”Language models are unsupervised multitask learn- ers,” OpenAI blog, vol","work_id":"f0bb251e-04ab-434f-98a5-128a671e70a0","year":2019},{"cited_arxiv_id":"2309.16609","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Qwen Technical Report","work_id":"bb1fd52f-6b2f-437c-9516-37bdf6eb9be8","year":2023}],"snapshot_sha256":"f0481316c66f551c336e9fd7a8f71926702a22696cd01431f9b788cb2643d180"},"source":{"id":"2605.15436","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T14:54:06.072337Z","id":"d7c379ca-341e-4e43-9379-7edb5136c2fd","model_set":{"reader":"grok-4.3"},"one_line_summary":"Analysis of 144 task-model pairs finds mathematical reasoning produces the highest attention entropy in all architectures while decoder models show significantly higher sparsity than encoders.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Mathematical reasoning produces the highest attention entropy across language model architectures.","strongest_claim":"Our analysis of 144 task-model combinations demonstrates that mathematical reasoning consistently produces the highest attention entropy across all architectures, while decoder models exhibit significantly higher sparsity patterns compared to encoder models.","weakest_assumption":"The twelve cognitive task categories and the chosen measurement definitions (final activation values, attention entropy, sparsity) are assumed to capture meaningful and comparable computational differences without substantial confounding from task formulation or model-specific tokenization effects."}},"verdict_id":"d7c379ca-341e-4e43-9379-7edb5136c2fd"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:9fda8e08b1fe6d06a2c6c1310540e334a9f1328bd89af1c9785dbda03af90418","target":"record","created_at":"2026-05-20T00:00:58Z","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":"323ee04d9b2ad9d88a7635b939cf925e81804fdeac22020879fc7707e7109867","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T21:31:19Z","title_canon_sha256":"45db951a6c79865d598c4d8bd1179df722501ed8c78f2d4a3dec3ec011210a1c"},"schema_version":"1.0","source":{"id":"2605.15436","kind":"arxiv","version":1}},"canonical_sha256":"acbe595fdcb1e08a47255dc5667b2c914dacfcd6cb65d6dd1705b2e6f51d185a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"acbe595fdcb1e08a47255dc5667b2c914dacfcd6cb65d6dd1705b2e6f51d185a","first_computed_at":"2026-05-20T00:00:58.510920Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:00:58.510920Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"M3LWu0kAO6aPsrK+WB+BnGrCqYQERPx1u2i/nOtNDkwH/sHAY/kOP0vJqneyjpQS7sRoFUC3fKPsq94/XBGWDQ==","signature_status":"signed_v1","signed_at":"2026-05-20T00:00:58.511767Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.15436","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:9fda8e08b1fe6d06a2c6c1310540e334a9f1328bd89af1c9785dbda03af90418","sha256:dfa1d7a17ed1306df020a38991b7dbec6944faa7904bf3e1f316c56988bf84d9"],"state_sha256":"6ce0d6b639524cd7e5c4b17951764b9133597dfa160bb0f689f7c5b10053e432"}