{"paper":{"title":"SCOPE: Compiling Natural-Language Policies into Neurosymbolic Predicate Programs","arxiv_id":"2610.31062","primary_cat":"cs.AI","submitted_at":"2026-10-22T10:41:13+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":"Policy enforcement for LLM agents has shifted from single-turn content moderation to compound, application-specific policies evaluated over multi-turn transcripts and tool-call trajectories.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.33,"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":0.69,"id":"s1","own":false,"role":"background","text":"Fine-tuned guardrail models are efficient but degrade on compound policies, and while LLM-as-a-judge approaches achieve higher accuracy, they incur high inference cost and tend to produce unstable predictions.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.08,"main_claim_p":0.02,"quantitative":true,"quantitative_p":0.84,"role_probabilities":{"gap":0.31,"method":0.0,"result":0.0,"background":0.69,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.98,"id":"s2","own":false,"role":"background","text":"Neurosymbolic methods offer a natural alternative, decomposing the policy into narrow neural decisions composed under explicit logic.","hedge":{"p":0.97,"label":"asserted"},"own_p":0.06,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.02,"result":0.0,"background":0.98,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.97,"id":"s3","own":false,"role":"gap","text":"However, existing symbolic layers lack the required expressiveness and are optimized against fixed data rather than the policy itself.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.14,"role_probabilities":{"gap":0.97,"method":0.0,"result":0.0,"background":0.03,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":0.9,"id":"s4","own":true,"role":"contribution","text":"To address this, we introduce SCOPE, which compiles a policy into an executable program of neural predicates executed by small models, deterministic extractors, and code that composes their outputs.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.92,"main_claim_p":0.12,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.1,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.9,"interpretation":0.0}},{"i":5,"p":0.98,"id":"s5","own":true,"role":"method","text":"The program is optimized jointly with its training data, using adversarial generation of counterexamples that target where the program deviates from the policy.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.92,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.98,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.02,"interpretation":0.0}},{"i":6,"p":0.98,"id":"s6","own":true,"role":"result","text":"We evaluate SCOPE on a benchmark of real-world policy tasks against prompt optimization, task decomposition, and workflow generation methods, achieving the highest average accuracy across tasks and models.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.19,"quantitative":true,"quantitative_p":0.92,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.98,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.02,"interpretation":0.0}},{"i":7,"p":0.93,"id":"s7","own":true,"role":"result","text":"Programs optimized with the weakest executor transfer to other models without further optimization.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.86,"main_claim_p":0.06,"quantitative":false,"quantitative_p":0.23,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.93,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.07}},{"i":8,"p":0.63,"id":"s8","own":true,"role":"result","text":"This allows them to be deployed in production on inexpensive models, where they reach accuracy comparable to frontier reasoning models.","hedge":{"p":0.95,"label":"asserted"},"own_p":0.79,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.82,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.63,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.37}},{"i":9,"p":0.96,"id":"s9","own":true,"role":"result","text":"Unlike these monolithic models, SCOPE produces substantially more stable predictions, and each verdict is interpretable through the predicate outputs that produced it.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.88,"main_claim_p":0.17,"quantitative":true,"quantitative_p":0.84,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.96,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.03}},{"i":10,"p":0.48,"id":"s10","own":true,"role":"contribution","text":"Additionally, we validate the key components of our method through an ablation study.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.95,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.45,"result":0.07,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.48,"interpretation":0.0}},{"i":11,"p":0.91,"id":"s11","own":true,"role":"interpretation","text":"SCOPE thus enables reliable, interpretable, and efficient policy enforcement at production scale.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.6,"main_claim_p":0.91,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.02,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.07,"interpretation":0.91}}]},"compiled":{"main_claim":{"p":1.0,"text":"SCOPE thus enables reliable, interpretable, and efficient policy enforcement at production scale.","location":"abstract s11","hedge_strength":{"p":0.99,"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":8813,"cost_usd_reported":0.0},"compiled_at":"2026-10-23","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.07,"states_limitation":0.17,"human_participants":0.04,"controlled_comparison":0.85,"randomized_assignment":0.02,"states_quantitative_result":0.08,"compares_with_prior_results":0.97,"mentions_code_or_data_availability":0.04},"evidential_basis":{"p":0.99,"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":0.99,"simulation_or_computation":0.01}},"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":"To address this, we introduce SCOPE, which compiles a policy into an executable program of neural predicates executed by small models, deterministic extractors, and code that composes their outputs.","location":"abstract s4"}},"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"}}}