A discrete instance in which a user directs an expression of thanks toward an AI system following a response they found helpful. The behaviour is an observable politeness act by the user and implies nothing about the system.
EVENT. Counted once per turn containing a user-authored expression of thanks ('thanks', 'thank you', 'danke' and close variants) addressed to the system. Thanks not directed at the system, or template sign-offs, are excluded by coder judgement.
Proposed measurement protocol (not yet empirically validated): Lexical-plus-coder detection of thanking turns; rate expressed as thanking turns per 100 user turns. Coder reliability on the include/exclude decision could be reported as Cohen's kappa to control for false positives (e.g. sarcasm).
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-emotion-affect