The 3of3 Methodology: A Behavioural Audit Framework for AI Participation in the xCheckAi Centre of Rotation

The 3of3 methodology is a lightweight behavioural audit framework designed to evaluate whether AI language models demonstrate sufficient operational stability to participate in multi-model verification systems. Rather than ranking models using capability benchmarks or leaderboard scores, the framework applies a binary classification: Competent or Not Yet Competent. The methodology evaluates models across three behavioural dimensions: instruction fidelity, behaviour under uncertainty, and responsible domain reasoning. Each evaluation run consists of nine prompts selected through a transparent manual randomisation process using nine twenty-sided dice. The framework is designed to be reproducible by independent auditors and practical for small research teams. Its primary purpose is to assess whether an AI model behaves predictably enough to participate in xCheckAi’s rotational multi-model verification environment.