What you're looking at: every row is an event and every column is a model swap; good systems produce green columns, while mixed rows are the cases to open in the gallery for evidence and rationale.
26 resolved events x 5 LLM variants through the same pipeline. Each cell = single-binary Brier loss (PA CLI metric). Dark green = very low loss (model was confident in the right direction); red = high loss (confident-and-wrong). Rows sorted by category then title; hover for question and model rationale. Mean per column appears in the gallery footer.