{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PVQVSZVIMTXHASXHXA4SVEX6GY","short_pith_number":"pith:PVQVSZVI","schema_version":"1.0","canonical_sha256":"7d615966a864ee704ae7b8392a92fe3631ad0af674c7e1179108654fe5add4b0","source":{"kind":"arxiv","id":"2602.11619","version":2},"attestation_state":"computed","paper":{"title":"When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aman Mehta","submitted_at":"2026-02-12T06:15:14Z","abstract_excerpt":"Running the same LLM agent on identical inputs yields 2.3-4.2 distinct action sequences per 10 runs; this behavioral variance constitutes a training-free, black-box uncertainty signal that instantiates selective classification and distribution-free calibration for agentic systems. Across 8,000 runs of four models on 200 HotpotQA questions, consistent tasks (at most 2 unique paths) achieve 82-87% accuracy while inconsistent tasks (4 or more paths) achieve 41-65%, a gap that survives controls for task difficulty. Divergence concentrates at step 2 (50.5% of Llama tasks), and consistency metrics d"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2602.11619","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-02-12T06:15:14Z","cross_cats_sorted":[],"title_canon_sha256":"7f48dd60d781344b1abdf9d2e3cdb00f151aa63bd305613b1d4e2b19ade77833","abstract_canon_sha256":"8f845ecbda58bf3b2f81bebad2e613b0b477651dd6308cbcbcdfab794cd5bad2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-16T00:21:45.284302Z","signature_b64":"qYGDe6xGHEsCrygQxOaoLbIenngYEpD8ZfmvgT0edcz3WmkLV//MftjmbiC8p7TMoBkcD6/tjpCWQVMIFwbTDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7d615966a864ee704ae7b8392a92fe3631ad0af674c7e1179108654fe5add4b0","last_reissued_at":"2026-07-16T00:21:45.283421Z","signature_status":"signed_v1","first_computed_at":"2026-07-16T00:21:45.283421Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aman Mehta","submitted_at":"2026-02-12T06:15:14Z","abstract_excerpt":"Running the same LLM agent on identical inputs yields 2.3-4.2 distinct action sequences per 10 runs; this behavioral variance constitutes a training-free, black-box uncertainty signal that instantiates selective classification and distribution-free calibration for agentic systems. Across 8,000 runs of four models on 200 HotpotQA questions, consistent tasks (at most 2 unique paths) achieve 82-87% accuracy while inconsistent tasks (4 or more paths) achieve 41-65%, a gap that survives controls for task difficulty. Divergence concentrates at step 2 (50.5% of Llama tasks), and consistency metrics d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.11619","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.11619/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2602.11619","created_at":"2026-07-16T00:21:45.283849+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.11619v2","created_at":"2026-07-16T00:21:45.283849+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.11619","created_at":"2026-07-16T00:21:45.283849+00:00"},{"alias_kind":"pith_short_12","alias_value":"PVQVSZVIMTXH","created_at":"2026-07-16T00:21:45.283849+00:00"},{"alias_kind":"pith_short_16","alias_value":"PVQVSZVIMTXHASXH","created_at":"2026-07-16T00:21:45.283849+00:00"},{"alias_kind":"pith_short_8","alias_value":"PVQVSZVI","created_at":"2026-07-16T00:21:45.283849+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"2605.28840","citing_title":"How Consistent Are LLM Agents? Measuring Behavioral Reproducibility in Multi-Step Tool-Calling Pipelines","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2606.22953","citing_title":"Plans Don't Persist: Why Context Management Is Load Bearing for LLM Agents","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2606.22936","citing_title":"When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agents","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20662","citing_title":"Confidence Laundering in Agent Systems: Why Uncertainty Needs a Latent Carrier","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2604.07595","citing_title":"ROZA Graphs: Self-Improving Near-Deterministic RAG through Evidence-Centric Feedback","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY","json":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY.json","graph_json":"https://pith.science/api/pith-number/PVQVSZVIMTXHASXHXA4SVEX6GY/graph.json","events_json":"https://pith.science/api/pith-number/PVQVSZVIMTXHASXHXA4SVEX6GY/events.json","paper":"https://pith.science/paper/PVQVSZVI"},"agent_actions":{"view_html":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY","download_json":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY.json","view_paper":"https://pith.science/paper/PVQVSZVI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.11619&json=true","fetch_graph":"https://pith.science/api/pith-number/PVQVSZVIMTXHASXHXA4SVEX6GY/graph.json","fetch_events":"https://pith.science/api/pith-number/PVQVSZVIMTXHASXHXA4SVEX6GY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY/action/storage_attestation","attest_author":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY/action/author_attestation","sign_citation":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY/action/citation_signature","submit_replication":"https://pith.science/pith/PVQVSZVIMTXHASXHXA4SVEX6GY/action/replication_record"}},"created_at":"2026-07-16T00:21:45.283849+00:00","updated_at":"2026-07-16T00:21:45.283849+00:00"}