Constructive Enquiry and Acquired Model Extension

Technical companion to the Textbook: The Aligned Signal by Kimberley Asher Abstract An internally consistent diagnostic system can admit an incorrect action when its model family omits the actual mechanism. We study two bounded responses: constructing additional enquiry to resolve represented action disagreement, and acquiring a response-model expression from independently audited admission failures. The first study improves main completion from 7,353 to 8,824 but makes 1,389 incorrect admissions outside its AND/XOR library. The second selects an expression from a fixed 26-program grammar using four audited calibration failures, validates it on separate measurements and retains it for later diagnosis. In 43,320 main episodes per arm, retention completes 12,994 cases versus 9,022 after removal and reduces incorrect admissions from 3,974 to zero. Exact restoration recovers the inherited result. At generous runtime resources, inherited and conventional expanded diagnosis complete all 8,664 eligible opportunities; the conventional method substantially outperforms inheritance under tight budgets. Two further omitted mechanisms cause 2,405 incorrect inherited admissions in 2,888 stress cases. The results support acquired model-library extension and changed subsequent diagnostic construction under fixed authority. They do not establish expansion of the expressive language, improvement of the outer learner, universal model coverage or general meta-RSI.