{ "method": "Previously exposed LLM questions; regression, not independent TEST.", "graph_freeze_sha256": "e5c6c01277478b49e10b0ea3daee991b8a20d895b5fb435506a01d8a941535ef", "audit_sha256": "d5922db25b6d40cb7e5dc7df7990e93a3caee263a6e747ae910ab340da340023", "generation_files": { "/home/janko/DP/zp-agent/tools/answer_evaluation/evaluate_graph_answers.py": "2553a0d17e164922b02046416fdf0d5df4496dc443466d8d9a1f4be86415b318", "/home/janko/DP/zp-agent/tools/answer_evaluation/graph_answer_prompt.txt": "c7104ffb08b5b384f38d48fbdad40ccd40de61beace984d8a118cf12bd229860", "/home/janko/DP/zp-agent/tools/answer_evaluation/answer_format.py": "86a27b4aebc6a5077e13794b81957a1cf063a1f3e070c583120ba53c1788d2f8", "/home/janko/DP/zp-agent/tools/answer_evaluation/context_policy.py": "ee3c423772800d3e7368fff9b48bcd464a98a741b87a0c79158398ed22a3dc00", "/home/janko/DP/zp-agent/tools/answer_evaluation/checked_generation.py": "e82b95314b8c0ae9ba82353b98876da7b2a4dce464eb21129ad904bf58b8ef5e", "tools/rag/evaluation/rag_runner.py": "031abf7d5b1a026eb199aa2a579bfb8baaa01175279d0018a364bfe50b76c3c1" }, "model": "model120-fast", "temperature": 0, "workers": 3, "dev_ids": [ "wide-001", "wide-002", "wide-041", "wide-042", "wide-080", "wide-081", "wide-082", "wide-104", "wide-113", "wide-114", "wide-116", "wide-119", "wide-129", "wide-137", "wide-138", "wide-139", "wide-141", "wide-142", "wide-143", "wide-145", "wide-155", "wide-159", "wide-161", "wide-163", "wide-165", "wide-169", "wide-170", "wide-178", "wide-181", "wide-193", "wide-194", "wide-209", "wide-217", "wide-218", "wide-232", "wide-236", "wide-239", "wide-249", "wide-250", "wide-283", "wide-289", "wide-290" ], "test_ids": [ "wide-003", "wide-004", "wide-005", "wide-006", "wide-007", "wide-008", "wide-009", "wide-010", "wide-011", "wide-012", "wide-013", "wide-014", "wide-015", "wide-016", "wide-017", "wide-018", "wide-019", "wide-020", "wide-021", "wide-022", "wide-023", "wide-024", "wide-025", "wide-026", "wide-027", "wide-028", "wide-029", "wide-030", "wide-031", "wide-032", "wide-033", "wide-034", "wide-035", "wide-036", "wide-037", "wide-038", "wide-039", "wide-040", "wide-043", "wide-044", "wide-045", "wide-046", "wide-047", "wide-048", "wide-049", "wide-050", "wide-051", "wide-052", "wide-053", "wide-054", "wide-055", "wide-056", "wide-057", "wide-058", "wide-059", "wide-060", "wide-061", "wide-062", "wide-063", "wide-064", "wide-065", "wide-066", "wide-067", "wide-068", "wide-069", "wide-070", "wide-071", "wide-072", "wide-073", "wide-074", "wide-075", "wide-076", "wide-077", "wide-078", "wide-079", "wide-080", "wide-083", "wide-084", "wide-085", "wide-086", "wide-087", "wide-088", "wide-089", "wide-090", "wide-091", "wide-092", "wide-093", "wide-094", "wide-095", "wide-096", "wide-097", "wide-098", "wide-099", "wide-100", "wide-101", "wide-102", "wide-103", "wide-104", "wide-105", "wide-106", "wide-107", "wide-108", "wide-109", "wide-110", "wide-111", "wide-112", "wide-115", "wide-116", "wide-117", "wide-118", "wide-119", "wide-120", "wide-121", "wide-122", "wide-123", "wide-124", "wide-125", "wide-126", "wide-127", "wide-128", "wide-129", "wide-130", "wide-131", "wide-132", "wide-133", "wide-134", "wide-135", "wide-136", "wide-139", "wide-140", "wide-141", "wide-142", "wide-143", "wide-144", "wide-145", "wide-146", "wide-147", "wide-148", "wide-149", "wide-150", "wide-151", "wide-152", "wide-153", "wide-154", "wide-155", "wide-156", "wide-157", "wide-158", "wide-159", "wide-160", "wide-161", "wide-162", "wide-163", "wide-164", "wide-165", "wide-166", "wide-167", "wide-168", "wide-171", "wide-172", "wide-173", "wide-174", "wide-175", "wide-176", "wide-177", "wide-178", "wide-179", "wide-180", "wide-181", "wide-182", "wide-183", "wide-184", "wide-185", "wide-186", "wide-187", "wide-188", "wide-189", "wide-190", "wide-191", "wide-192", "wide-195", "wide-196", "wide-197", "wide-198", "wide-199", "wide-200", "wide-201", "wide-202", "wide-203", "wide-204", "wide-205", "wide-206", "wide-207", "wide-208", "wide-209", "wide-210", "wide-211", "wide-212", "wide-213", "wide-214", "wide-215", "wide-216", "wide-219", "wide-220", "wide-221", "wide-222", "wide-223", "wide-224", "wide-225", "wide-226", "wide-227", "wide-228", "wide-229", "wide-230", "wide-231", "wide-232", "wide-233", "wide-234", "wide-235", "wide-236", "wide-237", "wide-238", "wide-239", "wide-240", "wide-241", "wide-242", "wide-243", "wide-244", "wide-245", "wide-246", "wide-247", "wide-248", "wide-251", "wide-252", "wide-253", "wide-254", "wide-255", "wide-256", "wide-257", "wide-258", "wide-259", "wide-260", "wide-261", "wide-262", "wide-263", "wide-264", "wide-265", "wide-266", "wide-267", "wide-268", "wide-269", "wide-270", "wide-271", "wide-272", "wide-273", "wide-274", "wide-275", "wide-276", "wide-277", "wide-278", "wide-279", "wide-280", "wide-281", "wide-282", "wide-283", "wide-284", "wide-285", "wide-286", "wide-287", "wide-288", "wide-291", "wide-292", "wide-293", "wide-294", "wide-295", "wide-296", "wide-297", "wide-298", "wide-299", "wide-300", "wide-301", "wide-302", "wide-303", "wide-304", "wide-305", "wide-306", "wide-307", "wide-308", "wide-309", "wide-310", "wide-311", "wide-312", "wide-313", "wide-314", "wide-315", "wide-316", "wide-317", "wide-318", "wide-319", "wide-320" ], "dev_count": 42, "test_count": 300, "target_accuracy": 0.92, "selected_split": "dev", "selected_ids": [ "wide-001", "wide-002", "wide-041", "wide-042", "wide-080", "wide-081", "wide-082", "wide-104", "wide-113", "wide-114", "wide-116", "wide-119", "wide-129", "wide-137", "wide-138", "wide-139", "wide-141", "wide-142", "wide-143", "wide-145", "wide-155", "wide-159", "wide-161", "wide-163", "wide-165", "wide-169", "wide-170", "wide-178", "wide-181", "wide-193", "wide-194", "wide-209", "wide-217", "wide-218", "wide-232", "wide-236", "wide-239", "wide-249", "wide-250", "wide-283", "wide-289", "wide-290" ], "selected_count": 42, "selected_dev_overlap": 42, "corrective_request_limit": 1, "format_repair_limit": 1, "repair_policy": "At most one total corrective request for format, citations or retrieval-context violations. Same question/context, no gold; retain all attempts.", "dev_results_sha256": null, "policy": "Evaluate every declared regression case, including DEV overlap; retain all failures and stop tuning after this freeze.", "selected_categories": { "shared_keyword_pair": 2, "same_academic_year_pair": 3, "same_start_year_pair": 3, "person_title_year": 5, "person_keyword_title": 12, "document_author_category_multi_hop": 4, "person_multiple_categories": 3, "candidate_set_aggregation": 5, "conjunctive_disambiguation": 3, "negative_lookup": 2 }, "limitations": [ "All questions and previous model outputs are known; this is historical regression.", "Harness supplies graph context; native OpenWebUI tool choice and browser UI are outside this experiment.", "Checks structured factual claims, their source support/citations and required prose values; not a general semantic judge of arbitrary prose." ] }