Corrective Comparator Experiment — DCI by Brand Pair

Cross-model mean Dimensional Collapse Index (DCI) for three focal brands under two comparator conditions: the default Anglophone-template comparator and a structurally corrective same-class comparator. Lower DCI indicates more dimensional richness preserved. VkusVill shows a +7.4 point increase in collapse when paired with the structurally correct Trader Joe's vs. Whole Foods, revealing the comparator trap.

Three-row table showing focal brands VkusVill, Calbee, and Roshen with their control comparators (Whole Foods, Frito-Lay, Cadbury), control DCI values (28.9, 38.3, 36.3), corrective comparators (Trader Joe's, Koikeya, Hershey), corrective DCI values (36.3, 39.0, 36.8), and delta (+7.4, +0.7, +0.4). VkusVill's delta is the largest single-pair comparator effect observed in any R15 condition.

(Please use a modern browser to see the interactive version of this visualization)

DCI = Dimensional Collapse Index. Lower = more dimensional richness preserved. Data from R15 Run 10 (126 API calls, six brand pairs, seven models, three runs per cell). Delta is corrective DCI minus control DCI; positive values indicate the corrective comparator made collapse worse.

Chart: Dmitry ZharnikovSource: R15: Dimensional Collapse in AI-Mediated Brand Perception (Zharnikov, 2026v)Get the dataCreated with Datawrapper