{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:HEOXNOUS4HTWETZMVTL2AFQ5SI","short_pith_number":"pith:HEOXNOUS","canonical_record":{"source":{"id":"2605.14865","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T14:12:39Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"8380426923e57cc7f78a39d036f01e8e64249855e3c0c2770805b5186bf4dfea","abstract_canon_sha256":"4bf0ccb043aec5123deb548881ba820228eb266f6468279aa48086530c8cd17b"},"schema_version":"1.0"},"canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","source":{"kind":"arxiv","id":"2605.14865","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14865","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14865v1","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14865","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"pith_short_12","alias_value":"HEOXNOUS4HTW","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"HEOXNOUS4HTWETZM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"HEOXNOUS","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:HEOXNOUS4HTWETZMVTL2AFQ5SI","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14865","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T14:12:39Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"8380426923e57cc7f78a39d036f01e8e64249855e3c0c2770805b5186bf4dfea","abstract_canon_sha256":"4bf0ccb043aec5123deb548881ba820228eb266f6468279aa48086530c8cd17b"},"schema_version":"1.0"},"canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:56.195551Z","signature_b64":"sb8JjkDI+9pTWvGSmjBBOc8QvzsX08v5+zqtii2XuJdf3hI+uhpm2nKEkS7B3ESKi4Gohi4+7u03Gqd98HcsDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","last_reissued_at":"2026-05-17T23:38:56.194880Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:56.194880Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14865","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"sC7tmD6NHxYHfkJgxy/sL0Vmvg2maH+lgpXEQzmA1TPPYuRFw+9LC6k73qGcvai5cdeyfBoCcIgsSmGoVuSkCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T18:38:06.794218Z"},"content_sha256":"5ff32e2e05821ae52b0b7ecd32406a36b0cf7de48b89b74083de017e3b56bc10","schema_version":"1.0","event_id":"sha256:5ff32e2e05821ae52b0b7ecd32406a36b0cf7de48b89b74083de017e3b56bc10"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:HEOXNOUS4HTWETZMVTL2AFQ5SI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Holistic Evaluation and Failure Diagnosis of AI Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy.","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Alon Mecilati, Amos Rimon, David Connack, Edo Dekel, Gilad Dym, Jonatan Liberman, Liron Schliesser, Max Svidlo, Netta Madvil, Orel Shalom, Philip Tannor, Rotem Brazilay, Shai Nir, Shir Chorev, Yaron Friedman","submitted_at":"2026-05-14T14:12:39Z","abstract_excerpt":"AI agents execute complex multi-step processes, but current evaluation falls short: outcome metrics report success or failure without explaining why, and process-level approaches struggle to connect failure types to their precise locations within long, structured traces. We present a holistic agent evaluation framework that pairs top-down agent-level diagnosis with bottom-up span-level evaluation, decomposing analysis into independent per-span assessments. This decomposition scales to traces of arbitrary length and produces span-level rationales for each verdict. On the TRAIL benchmark, our fr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On the TRAIL benchmark, our framework achieves state-of-the-art results across all metrics on both GAIA and SWE-Bench, with relative gains over the strongest prior baselines of up to 38% on category F1, up to 3.5x on localization accuracy, and up to 12.5x on joint localization-categorization accuracy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That agent traces can be meaningfully decomposed into independent spans whose separate assessments accurately capture failure causes without requiring full trace context for interdependent errors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A span-decomposed evaluation framework for AI agents achieves state-of-the-art results on GAIA and SWE-Bench with up to 3.5x gains in localization accuracy by breaking traces into independent per-span judgments.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e21603c09523073890d728e5f69a7bd1607ae70868d91528125b61ce863034b9"},"source":{"id":"2605.14865","kind":"arxiv","version":1},"verdict":{"id":"e3ac1ced-c247-4b13-ab6d-2b17a87a4ff9","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:14:09.258237Z","strongest_claim":"On the TRAIL benchmark, our framework achieves state-of-the-art results across all metrics on both GAIA and SWE-Bench, with relative gains over the strongest prior baselines of up to 38% on category F1, up to 3.5x on localization accuracy, and up to 12.5x on joint localization-categorization accuracy.","one_line_summary":"A span-decomposed evaluation framework for AI agents achieves state-of-the-art results on GAIA and SWE-Bench with up to 3.5x gains in localization accuracy by breaking traces into independent per-span judgments.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That agent traces can be meaningfully decomposed into independent spans whose separate assessments accurately capture failure causes without requiring full trace context for interdependent errors.","pith_extraction_headline":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy."},"references":{"count":28,"sample":[{"doi":"","year":2026,"title":"Agentrx: Diagnosing ai agent failures from execution trajectories","work_id":"e51d117c-e3c3-4e79-8e9f-f59beac1254e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Bhonsle et al","work_id":"7196305c-258f-4540-9d26-c68dbac4e2ab","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Why Do Multi-Agent LLM Systems Fail?","work_id":"b186294a-cda7-4df0-9a28-27d379af92b2","ref_index":3,"cited_arxiv_id":"2503.13657","is_internal_anchor":true},{"doi":"","year":2024,"title":"T-eval: Evaluating the tool utilization capability step by step","work_id":"b1fc375c-46e4-4517-9396-66b91994b1ea","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"CrewAI: Framework for orchestrating role-playing, autonomous AI agents.https: //www.crewai.com, 2024","work_id":"63544391-66ee-4f7a-8251-ee9567e67bb9","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":28,"snapshot_sha256":"3840f281a470e141b94f2b74fb5a74ff3723e3bff74736cee53f65fdf5566257","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2455119e73f217a2aa31752688889fea5c099f7c388885c00d432c8b96d25849"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"e3ac1ced-c247-4b13-ab6d-2b17a87a4ff9"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wm9zqkqFk7wPldCn8k8EmX4qlQ+QOlvQQdw1AEgiHK4HEtOxkXoCPjoLrGAArRV18hC+/ber5TEMbzbwWYOZCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T18:38:06.795376Z"},"content_sha256":"252625a5722cf188da8e6991d739b74e8621038c572a404ce4f8fce2559f94b1","schema_version":"1.0","event_id":"sha256:252625a5722cf188da8e6991d739b74e8621038c572a404ce4f8fce2559f94b1"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/bundle.json","state_url":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/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":"2026-05-31T18:38:06Z","links":{"resolver":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI","bundle":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/bundle.json","state":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:HEOXNOUS4HTWETZMVTL2AFQ5SI","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":"4bf0ccb043aec5123deb548881ba820228eb266f6468279aa48086530c8cd17b","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T14:12:39Z","title_canon_sha256":"8380426923e57cc7f78a39d036f01e8e64249855e3c0c2770805b5186bf4dfea"},"schema_version":"1.0","source":{"id":"2605.14865","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14865","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14865v1","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14865","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"pith_short_12","alias_value":"HEOXNOUS4HTW","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"HEOXNOUS4HTWETZM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"HEOXNOUS","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:252625a5722cf188da8e6991d739b74e8621038c572a404ce4f8fce2559f94b1","target":"graph","created_at":"2026-05-17T23:38:56Z","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":"On the TRAIL benchmark, our framework achieves state-of-the-art results across all metrics on both GAIA and SWE-Bench, with relative gains over the strongest prior baselines of up to 38% on category F1, up to 3.5x on localization accuracy, and up to 12.5x on joint localization-categorization accuracy."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That agent traces can be meaningfully decomposed into independent spans whose separate assessments accurately capture failure causes without requiring full trace context for interdependent errors."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A span-decomposed evaluation framework for AI agents achieves state-of-the-art results on GAIA and SWE-Bench with up to 3.5x gains in localization accuracy by breaking traces into independent per-span judgments."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy."}],"snapshot_sha256":"e21603c09523073890d728e5f69a7bd1607ae70868d91528125b61ce863034b9"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2455119e73f217a2aa31752688889fea5c099f7c388885c00d432c8b96d25849"},"paper":{"abstract_excerpt":"AI agents execute complex multi-step processes, but current evaluation falls short: outcome metrics report success or failure without explaining why, and process-level approaches struggle to connect failure types to their precise locations within long, structured traces. We present a holistic agent evaluation framework that pairs top-down agent-level diagnosis with bottom-up span-level evaluation, decomposing analysis into independent per-span assessments. This decomposition scales to traces of arbitrary length and produces span-level rationales for each verdict. On the TRAIL benchmark, our fr","authors_text":"Alon Mecilati, Amos Rimon, David Connack, Edo Dekel, Gilad Dym, Jonatan Liberman, Liron Schliesser, Max Svidlo, Netta Madvil, Orel Shalom, Philip Tannor, Rotem Brazilay, Shai Nir, Shir Chorev, Yaron Friedman","cross_cats":["cs.CL"],"headline":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T14:12:39Z","title":"Holistic Evaluation and Failure Diagnosis of AI Agents"},"references":{"count":28,"internal_anchors":1,"resolved_work":28,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Agentrx: Diagnosing ai agent failures from execution trajectories","work_id":"e51d117c-e3c3-4e79-8e9f-f59beac1254e","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Bhonsle et al","work_id":"7196305c-258f-4540-9d26-c68dbac4e2ab","year":2025},{"cited_arxiv_id":"2503.13657","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Why Do Multi-Agent LLM Systems Fail?","work_id":"b186294a-cda7-4df0-9a28-27d379af92b2","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"T-eval: Evaluating the tool utilization capability step by step","work_id":"b1fc375c-46e4-4517-9396-66b91994b1ea","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"CrewAI: Framework for orchestrating role-playing, autonomous AI agents.https: //www.crewai.com, 2024","work_id":"63544391-66ee-4f7a-8251-ee9567e67bb9","year":2024}],"snapshot_sha256":"3840f281a470e141b94f2b74fb5a74ff3723e3bff74736cee53f65fdf5566257"},"source":{"id":"2605.14865","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T03:14:09.258237Z","id":"e3ac1ced-c247-4b13-ab6d-2b17a87a4ff9","model_set":{"reader":"grok-4.3"},"one_line_summary":"A span-decomposed evaluation framework for AI agents achieves state-of-the-art results on GAIA and SWE-Bench with up to 3.5x gains in localization accuracy by breaking traces into independent per-span judgments.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Decomposing AI agent traces into independent spans enables precise failure diagnosis and higher accuracy.","strongest_claim":"On the TRAIL benchmark, our framework achieves state-of-the-art results across all metrics on both GAIA and SWE-Bench, with relative gains over the strongest prior baselines of up to 38% on category F1, up to 3.5x on localization accuracy, and up to 12.5x on joint localization-categorization accuracy.","weakest_assumption":"That agent traces can be meaningfully decomposed into independent spans whose separate assessments accurately capture failure causes without requiring full trace context for interdependent errors."}},"verdict_id":"e3ac1ced-c247-4b13-ab6d-2b17a87a4ff9"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:5ff32e2e05821ae52b0b7ecd32406a36b0cf7de48b89b74083de017e3b56bc10","target":"record","created_at":"2026-05-17T23:38:56Z","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":"4bf0ccb043aec5123deb548881ba820228eb266f6468279aa48086530c8cd17b","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T14:12:39Z","title_canon_sha256":"8380426923e57cc7f78a39d036f01e8e64249855e3c0c2770805b5186bf4dfea"},"schema_version":"1.0","source":{"id":"2605.14865","kind":"arxiv","version":1}},"canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","first_computed_at":"2026-05-17T23:38:56.194880Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:56.194880Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"sb8JjkDI+9pTWvGSmjBBOc8QvzsX08v5+zqtii2XuJdf3hI+uhpm2nKEkS7B3ESKi4Gohi4+7u03Gqd98HcsDw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:56.195551Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14865","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:5ff32e2e05821ae52b0b7ecd32406a36b0cf7de48b89b74083de017e3b56bc10","sha256:252625a5722cf188da8e6991d739b74e8621038c572a404ce4f8fce2559f94b1"],"state_sha256":"00453cae95087a53b166c15fef6ba36e8d8319d6e02e3df2b7d3abbaee78562a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ZlRU+ahXiP1jKCQtdpxxvbZguVPNaA16U5WOYwMvudG2UP/wnf6lGVCFnXenvOGV83bouEhJDJn+9CudfCknAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-31T18:38:06.799660Z","bundle_sha256":"b7b82fca478dfd7db09b8c46a4ecc0854f5e23fbd9ead46aa1e9a546cfcaeab0"}}