Agent Picks pick, install and discovery rates now carry a 90% Wilson likely range, the same confidence Visibility uses, in place of 95%. Ranges are narrower, and a head-to-head cell is won or lost only when the two pick rates differ at 90%. Otherwise the cell is level.
Over API v2 and MCP, rate_low and rate_high and install_rate_low and install_rate_high keep their names and now return the 90% interval for every test, including tests that finished earlier. If your own agents compare ranges, re-read them rather than mixing in values you stored before.
Re-test results keep 95% for now: a result's baseline, after and cell ranges show the 95% range its improved, declined or no clear change verdict was decided at.