writing

The benchmark that didn't pick a winner

525 graded runs of rootcause across seven settings. Cost per solved incident varied 25x; pass rates of 77-84% showed no detected difference, though gaps under about 13-19 points would not have been detected.

25xspread in cost per solved incident

Getting a pricing model adopted at Reshet 13

A regression that found underpriced primetime ad slots, and the work with the commercial team until they set prices from it.