Inference as Training

This corrected preprint presents its research question, method, bounded result, and principal limitations in a standardized reader-facing format. Suppose a person misjudges the weight of a cup, corrects the movement, and lifts it more accurately on the next attempt. The later success may reflect more computation with an unchanged system, or it may reflect a change left by the first attempt. Those possibilities are often blurred when inference, adaptation, memory, and learning are discussed across neuroscience and artificial intelligence. This paper supplies an operational boundary. An inference episode becomes training-like when mismatch resolution changes a measurable state, that state persists beyond the immediate episode, and selectively removing it changes a later inference under matched conditions. It belongs to Micah Blumberg's governed research-paper corpus. This is a corrected preprint edition released under the Creative Commons Attribution 4.0 license.