Who is being messaged, who is being skipped, and which gate skipped them.

The disagreement rate says how often poll order and the selector differ. This says what is actually happening, which is the question you have while WATCHING a rollout - and the one that catches a bad one. A disagreement rate can look perfectly healthy while every push in the workspace is being dropped for a single reason.

Read skips_by_reason first. A spike in no_channel_token is an integration problem, a spike in frequency_cap is a budget problem, and a spike in premise_gone means people are coming back on their own, which is the outcome you wanted.