MITI 4.2.1 is the standard instrument for rating a clinician's fidelity to Motivational Interviewing. Alongside global scores, it tallies behavior counts — discrete, tally-able clinician utterances such as reflections, questions, affirmations, and MI-consistent or MI-inconsistent moves. These behavior counts are the instance-highlighting targets modelled here.
Each of the nine behaviour counts ships with a curated agent skill. Because the counts are tallies rather than ratings, the severity value records how clearly an utterance exemplifies its code (1 marginal, 2 clear, 3 prototypical) rather than a symptom intensity. Coding targets the clinician's utterances; the client's speech is left to the MISC client-language scale.
The skills operationalise the manual's own hard boundaries — simple versus complex reflection, giving information versus persuading, and persuade versus confront — with contrastive pairs on each line.