Every condition, one strip at a time
Condition explorer
Pick a stimulus-novelty tier and a format, and read each effect as one strip: the dark tick is the human benchmark with its shaded 95 percent interval, and each colored dot is one engine's effect size on identical scenarios (an engine is one model in one configuration). Flip to the by-engine view to see the same cells regrouped: every engine's distance from the human number, effect by effect. Flipping a facet moves the same dots to their new places.
View the plotted cells as a table
Benchmarks are pooled across the 13 product categories in the rehearsal pack; the per-category split is reported with the real data. Engine dots carry no interval by design: a model effect is a population quantity computed over the fully enumerated grid of orders, paraphrases, and personas, so its distance from the human number is a descriptive gap, never a test statistic. Hollow dots are cells the registered validity criteria held out of scoring (a ceiling baseline, or an effect that did not replicate in humans).