{"paper":{"title":"StateM: Reaching 95.3% Raw Accuracy, or a \\$15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling","arxiv_id":"2608.15089","primary_cat":"cs.AI","submitted_at":"2026-08-15T07:16:28+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":0.98,"id":"s0","own":false,"role":"background","text":"Long-horizon agents can fail even when their underlying models can solve the constituent steps.","hedge":{"p":0.77,"label":"asserted"},"own_p":0.12,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.02,"method":0.0,"result":0.0,"background":0.98,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.98,"id":"s1","own":false,"role":"background","text":"They may lose track of mutable state, fail to reactivate lessons from earlier executions, skip known procedures, or stop prematurely.","hedge":{"p":1.0,"label":"hedged_possibility"},"own_p":0.1,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.02,"method":0.0,"result":0.0,"background":0.98,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.56,"id":"s2","own":false,"role":"method","text":"We bet on harness scaling to improve the execution system around an agent without changing its model weights.","hedge":{"p":0.82,"label":"speculative"},"own_p":0.22,"main_claim_p":0.14,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.01,"method":0.56,"result":0.0,"background":0.03,"limitation":0.0,"future_work":0.0,"contribution":0.39,"interpretation":0.01}},{"i":3,"p":0.85,"id":"s3","own":true,"role":"contribution","text":"We introduce StateM, an agent-native runtime that organizes execution around durable states, phase-local context, checked transitions, recoverable runbooks, and versioned procedural practices that agents and users can inspect together.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.85,"main_claim_p":0.11,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.15,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.85,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"On Terminal-Bench 2.1, StateM raises GPT-5.5 xhigh to 92.1\\%, versus 83.1\\% reference and GPT-5.6 Sol Ultra at 91.9\\%.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.22,"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":5,"p":0.93,"id":"s5","own":true,"role":"result","text":"The runbook transfers unchanged to GPT-5.6.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.84,"main_claim_p":0.06,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.01,"result":0.93,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.05}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"With GPT-5.6 Sol xhigh, StateM reaches 95.3\\% raw accuracy across 445 trials and succeeds on all 89 tasks at least once.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.42,"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":1.0,"id":"s7","own":true,"role":"result","text":"The frozen profile raises GPT-5.6 Luna from 76.7 to 85.4\\%, above the 84.9\\% Sol xhigh reference.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.08,"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":8,"p":1.0,"id":"s8","own":true,"role":"result","text":"Using the same runtime, runbook structure, and golden rules, less than \\$38 of adaptation raises DeepSeek-V4 Flash from 82.7 to 88.1\\% under standard timeouts and to 89.1\\% on an 88-task common core.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"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":9,"p":1.0,"id":"s9","own":true,"role":"result","text":"Extending only the remaining latency-sensitive task matches the reported 88.8\\% GPT-5.6 Sol max result.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.86,"main_claim_p":0.07,"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":10,"p":1.0,"id":"s10","own":true,"role":"result","text":"Final-score API usage is about \\$15 versus \\$574.68 for the GPT reference; total DeepSeek expenditure is \\$52.22.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.08,"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":1.0,"id":"s11","own":true,"role":"result","text":"On BusinessBench, family-specific runbooks built on development sets yield held-out gains of 0.55 macro and 1.34 micro points; two mechanism-matched families improve by 10.04 points.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.06,"quantitative":true,"quantitative_p":0.98,"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":12,"p":0.98,"id":"s12","own":true,"role":"interpretation","text":"Concrete rules generalize when tasks share execution structure, while the control methodology applies broadly.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.7,"main_claim_p":0.26,"quantitative":false,"quantitative_p":0.07,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.02,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.98}},{"i":13,"p":0.46,"id":"s13","own":true,"role":"method","text":"StateM turns selected postmortem findings into persistent, executable preconditions and practices, making learned controls explicit and enforceable through stateful controls.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.84,"main_claim_p":0.27,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.46,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.39,"interpretation":0.15}},{"i":14,"p":0.74,"id":"s14","own":false,"role":"contribution","text":"Code at github.com/henryqin1997/statem.","hedge":{"p":0.69,"label":"unclear"},"own_p":0.48,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.11,"result":0.0,"background":0.06,"limitation":0.0,"future_work":0.09,"contribution":0.74,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.32,"text":"Concrete rules generalize when tasks share execution structure, while the control methodology applies broadly.","location":"abstract s12","hedge_strength":{"p":1.0,"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":11075,"cost_usd_reported":0.0},"compiled_at":"2026-09-23","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.05,"states_limitation":0.37,"human_participants":0.06,"controlled_comparison":0.94,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.98,"mentions_code_or_data_availability":0.97},"evidential_basis":{"p":0.97,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.97,"simulation_or_computation":0.03}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We introduce StateM, an agent-native runtime that organizes execution around durable states, phase-local context, checked transitions, recoverable runbooks, and versioned procedural practices that agents and users can inspect together.","location":"abstract s3"}},"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":15,"sentences_in_abstract":15},"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"}}}