Monitoring
Trust gates on broken data
Three conditions under which automation refuses to act, because the reading it would act on cannot be believed.
Automation is only as good as the number it reads. Three conditions mean the number cannot be read, and each one blocks any action that would be based on it.
A sample ratio mismatch means traffic did not split as configured, so the two groups are not comparable and no contrast between them holds. No traffic for a full day means nothing is being measured at all. And an interference finding means another running experiment is corrupting this reading, which is the one nothing else would catch.
That third gate is separate from the first rather than folded into it, because the two do not overlap the way they look like they should. A redirect interaction does surface as a split mismatch, but two experiments editing the same element produce a perfect split and a page combination nobody reviewed, which is invisible to the sample-ratio check and still not a reading worth acting on.
Trust gates only guard the significance path. The time-based stop is unconditional, because ending an experiment is always safe and leaving a broken one running past its planned end makes two problems out of one. When a gate blocks, the reason is stated in plain words rather than as a status code.