This document is released as a public technical research note. It is not part of the formal Architecture of Limitation corpus and should not be read as a preprint or theoretical extension. This report examines the Ministral local LM experimentation recorded in the run-comparison document, with `K0ᵥ43. yaml` used as the governing diagnostic constraint. The central finding is not that a local model can execute AoL/AoLOS perfectly. That claim is not supported by the experiment. The stronger finding is that a less sophisticated local substrate makes collapse signals more visible: linguistic capture under K1, formal overclassification under K0G, and framework over-assimilation under K0. Most importantly, the K0 runs exposed that unused stack elements named inside a contract, wrapper, or runner can become active influences in the reasoning field, even when mentioned only as exclusions. This shifts the design lesson from “add clearer prohibitions” to “declare only the active field. ” The experiment therefore succeeds not by arriving at the expected benchmark result, but by making an otherwise hidden AoLOS design weakness observable and correctable.
Franky Schaut (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: