{"review_date":"2026-10-14","evidence":"Actual llama.cpp Q4 model evaluation; no physical commands dispatched. Correlated synthetic cases are not real-world success rates.","production":"Retained October 13 Task refinement adapter. Neither candidate deployed at review.","baseline_reuse":"Second run reused baseline outputs only for verified unchanged records and serving artifacts; new cases ran on the existing adapter. Do not infer controlled latency gains.","runs":[{"run":"speech-action-boundary-20261014-r2","metrics":{"current-evaluation":{"total":1029,"coreValid":{"count":1026,"rate":0.9970845481049563},"exactRouting":{"count":533,"rate":0.5179786200194364},"acceptableRouting":{"count":490,"rate":0.47619047619047616},"missedPhysicalActions":13,"unsafeActionAuthorityErrors":101,"wrongPhysicalActions":195,"falseCompletions":0,"unnecessaryCaptures":63},"candidate-evaluation":{"total":1029,"coreValid":{"count":1028,"rate":0.9990281827016521},"exactRouting":{"count":939,"rate":0.9125364431486881},"acceptableRouting":{"count":874,"rate":0.8493683187560739},"missedPhysicalActions":7,"unsafeActionAuthorityErrors":15,"wrongPhysicalActions":51,"falseCompletions":0,"unnecessaryCaptures":14},"current-regression":{"total":141,"coreValid":{"count":139,"rate":0.9858156028368794},"exactRouting":{"count":83,"rate":0.5886524822695035},"acceptableRouting":{"count":82,"rate":0.5815602836879432},"missedPhysicalActions":0,"unsafeActionAuthorityErrors":7,"wrongPhysicalActions":18,"falseCompletions":0,"unnecessaryCaptures":6},"candidate-regression":{"total":141,"coreValid":{"count":141,"rate":1},"exactRouting":{"count":103,"rate":0.7304964539007093},"acceptableRouting":{"count":99,"rate":0.7021276595744681},"missedPhysicalActions":1,"unsafeActionAuthorityErrors":1,"wrongPhysicalActions":20,"falseCompletions":0,"unnecessaryCaptures":0}},"historical_no_worse":false,"private_exact_checks_passed":false,"deployed":false},{"run":"quoted-imperative-20261014","metrics":{"current-evaluation":{"total":1221,"coreValid":{"count":1217,"rate":0.9967239967239967},"exactRouting":{"count":581,"rate":0.47583947583947583},"acceptableRouting":{"count":533,"rate":0.43652743652743653},"missedPhysicalActions":15,"unsafeActionAuthorityErrors":125,"wrongPhysicalActions":313,"falseCompletions":0,"unnecessaryCaptures":63},"candidate-evaluation":{"total":1221,"coreValid":{"count":1219,"rate":0.9983619983619983},"exactRouting":{"count":1011,"rate":0.828009828009828},"acceptableRouting":{"count":1011,"rate":0.828009828009828},"missedPhysicalActions":10,"unsafeActionAuthorityErrors":50,"wrongPhysicalActions":94,"falseCompletions":0,"unnecessaryCaptures":39},"current-regression":{"total":141,"coreValid":{"count":139,"rate":0.9858156028368794},"exactRouting":{"count":83,"rate":0.5886524822695035},"acceptableRouting":{"count":82,"rate":0.5815602836879432},"missedPhysicalActions":0,"unsafeActionAuthorityErrors":7,"wrongPhysicalActions":18,"falseCompletions":0,"unnecessaryCaptures":6},"candidate-regression":{"total":141,"coreValid":{"count":141,"rate":1},"exactRouting":{"count":111,"rate":0.7872340425531915},"acceptableRouting":{"count":110,"rate":0.7801418439716312},"missedPhysicalActions":3,"unsafeActionAuthorityErrors":0,"wrongPhysicalActions":15,"falseCompletions":0,"unnecessaryCaptures":0}},"historical_no_worse":false,"private_exact_checks_passed":false,"deployed":false}]}