The tendency of an AI session to develop an increasing thematic pull toward its accumulated context as that context grows over the course of the session. It describes a longitudinal drift in topical concentration.
TREND. Tracked over the session timeline: the degree to which later turns remain within the topic established by accumulated earlier context, measured as change in topical concentration from early to late session windows. An upward trend indicates gravity.
Proposed measurement protocol (not yet empirically validated): Longitudinal topic-concentration index across binned session windows (e.g. embedding-based similarity of each turn to the running session centroid); slope over turns reported per session. Validated against coder topic-continuity ratings (ICC).
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-system-behavior