Personas react to impressions, not to individual stimuli

A persona reacts once per channel per tick, to the whole set of stimuli it saw (an impression), rather than once per stimulus. Seeing two things side by side is not the same as seeing each alone, and social proof — the mechanic that distinguishes the feed from the survey room — only operates through juxtaposition; serving stimuli one at a time reduces the feed to a slower survey room.

The cost difference is not marginal: at 2,000 personas over 30 ticks with an exposure budget of three, per-stimulus reactions mean roughly 180,000 tier-A calls per world against 60,000 for impressions, multiplied again by variants and replicates in a sweep.

Consequences

Exposure keeps its per-stimulus shape so recommender scoring, attention and drop-reasons are unchanged; exposures are grouped before reaching the agent, not flattened. A reaction records both the impression it saw and the subject_stimulus_id it is about, since purchase intent is directed at a specific proposition even when several things were on screen. A survey-room impression holds exactly one exposure, so both environments share one code path.