An interaction pattern in which AI output is edited by the user and returned for further processing across successive turns, producing progressive refinement of the artefact. It denotes an observable iterate-and-return turn sequence.
EVENT. Counted once per detected cycle in which user-modified content from a prior AI turn is resubmitted and the system produces a revised version. A run of k such cycles on one artefact is recorded as a chain of length k.
Proposed measurement protocol (not yet empirically validated): Turn-graph analysis of resubmission edges (content overlap between a user turn and the preceding AI turn above a similarity threshold); could be reported as cycle count and mean chain length per task. Coder validation of detected cycles via Cohen's kappa.
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-work-behavior