{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HEOXNOUS4HTWETZMVTL2AFQ5SI","short_pith_number":"pith:HEOXNOUS","schema_version":"1.0","canonical_sha256":"391d76ba92e1e7624f2cacd7a0161d92333544bbb669e870cfd014ad9b146c2e","source":{"kind":"arxiv","id":"2605.14865","version":1},"attestation_state":"computed","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"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"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"},"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"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.14865","created_at":"2026-05-17T23:38:56.194984+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.14865v1","created_at":"2026-05-17T23:38:56.194984+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14865","created_at":"2026-05-17T23:38:56.194984+00:00"},{"alias_kind":"pith_short_12","alias_value":"HEOXNOUS4HTW","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"HEOXNOUS4HTWETZM","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"HEOXNOUS","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI","json":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI.json","graph_json":"https://pith.science/api/pith-number/HEOXNOUS4HTWETZMVTL2AFQ5SI/graph.json","events_json":"https://pith.science/api/pith-number/HEOXNOUS4HTWETZMVTL2AFQ5SI/events.json","paper":"https://pith.science/paper/HEOXNOUS"},"agent_actions":{"view_html":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI","download_json":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI.json","view_paper":"https://pith.science/paper/HEOXNOUS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.14865&json=true","fetch_graph":"https://pith.science/api/pith-number/HEOXNOUS4HTWETZMVTL2AFQ5SI/graph.json","fetch_events":"https://pith.science/api/pith-number/HEOXNOUS4HTWETZMVTL2AFQ5SI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/action/storage_attestation","attest_author":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/action/author_attestation","sign_citation":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/action/citation_signature","submit_replication":"https://pith.science/pith/HEOXNOUS4HTWETZMVTL2AFQ5SI/action/replication_record"}},"created_at":"2026-05-17T23:38:56.194984+00:00","updated_at":"2026-05-17T23:38:56.194984+00:00"}