{"paper":{"title":"When to Intervene? An Empirical Study of Errors and Self-Recovery in LLM Agents","arxiv_id":"2610.36471","primary_cat":"cs.AI","submitted_at":"2026-10-27T01:26:53+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"LLM agents interact with Web services to perform tasks on users' behalf.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.04,"main_claim_p":0.01,"quantitative":false,"quantitative_p":0.02,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.99,"id":"s1","own":false,"role":"background","text":"Errors during these interactions can disrupt execution and prevent task completion, motivating external intervention.","hedge":{"p":0.9,"label":"asserted"},"own_p":0.05,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.02,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.0,"background":0.99,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.68,"id":"s2","own":false,"role":"gap","text":"However, agents may self-recover through continued execution, raising the question of when to intervene and when to wait for self-recovery.","hedge":{"p":0.97,"label":"hedged_possibility"},"own_p":0.05,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.68,"method":0.0,"result":0.0,"background":0.32,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"contribution","text":"We present an empirical study of errors and self-recovery in LLM agents.","hedge":{"p":0.95,"label":"asserted"},"own_p":0.94,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":1.0,"interpretation":0.0}},{"i":4,"p":0.8,"id":"s4","own":true,"role":"method","text":"We construct ReInBench, a dataset containing 1,259 agent traces and 1,403 error events across computer-use and Web tasks.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.02,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.8,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.2,"interpretation":0.0}},{"i":5,"p":0.81,"id":"s5","own":true,"role":"contribution","text":"Our study examines how often and how quickly agents self-recover, how self-recovery relates to subsequent execution and task completion, and whether self-recovery can be predicted when an error occurs.","hedge":{"p":0.54,"label":"asserted"},"own_p":0.72,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.01,"method":0.18,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.81,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"We find that 725 error events self-recover, often within a few steps, with self-recovery performance differing across Execution LLMs.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.11,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":0.95,"id":"s7","own":true,"role":"result","text":"Self-recovery can also help agents recover from subsequent errors of the same type and support overall task completion.","hedge":{"p":0.51,"label":"hedged_possibility"},"own_p":0.87,"main_claim_p":0.09,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.95,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.05}},{"i":8,"p":1.0,"id":"s8","own":false,"role":"interpretation","text":"Intervention should therefore prioritize errors unlikely to self-recover, while allowing agents to continue when self-recovery is likely.","hedge":{"p":0.83,"label":"asserted"},"own_p":0.33,"main_claim_p":0.68,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}},{"i":9,"p":0.61,"id":"s9","own":true,"role":"method","text":"Building on these empirical findings, we propose ReCoF, a method that predicts whether an error will self-recover to inform whether additional intervention is needed.","hedge":{"p":0.97,"label":"asserted"},"own_p":0.93,"main_claim_p":0.15,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.61,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.39,"interpretation":0.0}},{"i":10,"p":1.0,"id":"s10","own":true,"role":"result","text":"ReCoF achieves an average relative improvement of 9.02% in Macro-F1 over the strongest baseline on ReInBench.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.09,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":11,"p":0.83,"id":"s11","own":true,"role":"contribution","text":"Our code and data can be found at https://method-notes.github.io/ReCoF/.","hedge":{"p":0.96,"label":"asserted"},"own_p":0.55,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.02,"result":0.01,"background":0.01,"limitation":0.0,"future_work":0.13,"contribution":0.83,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.99,"text":"Intervention should therefore prioritize errors unlikely to self-recover, while allowing agents to continue when self-recovery is likely.","location":"abstract s8","hedge_strength":{"p":0.83,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":8783,"cost_usd_reported":0.0},"compiled_at":"2026-10-28","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.05,"states_limitation":0.1,"human_participants":0.12,"controlled_comparison":0.72,"randomized_assignment":0.01,"states_quantitative_result":0.99,"compares_with_prior_results":0.94,"mentions_code_or_data_availability":0.99},"evidential_basis":{"p":0.96,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.04,"observation_or_experiment":0.96,"simulation_or_computation":0.0}},"contribution_type":{"p":0.97,"value":"empirical study","probabilities":{"review":0.0,"method_or_tool":0.01,"empirical_study":0.97,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.02,"position_or_commentary":0.0}},"contribution_statement":{"p":0.73,"text":"Building on these empirical findings, we propose ReCoF, a method that predicts whether an error will self-recover to inform whether additional intervention is needed.","location":"abstract s9"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":12,"sentences_in_abstract":12},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}