AI Verification Methodology

Published in 2026 by Masonsoft Technology Ltd, the AI Verification Methodology is a practical framework for consistently evaluating AI output in proportion to what is actually at stake. It takes the Calibrated Thinking posture and gives it operational structure: not just the instruction to verify, but the means to do it well. The methodology distinguishes between verification as it typically happens, applied inconsistently under time pressure and without a shared language for what went wrong, and verification as it should operate: a consistent six-step habit applied at the point of work. The six steps cover a pre-submission audit of the prompt before it is sent, output classification across three sensitivity levels, plausibility checking, structured challenge using the three Calibrated Thinking questions, a decision gate with four possible outcomes, and a standardised verification record. The document is addressed in two sections: one for those with management responsibility, outlining the governance actions required before the employee process can operate effectively; and one for employees doing the work. A failure taxonomy defines four types, covering prompt failure, context failure, capability failure, and tool failure, each with a different remedy. An appendix provides the Calibrated Thinking quick-reference card.