When a channel goes quiet overnight or attribution suddenly stops reconciling, the first reaction is often 'were we throttled — should we switch channels now?'. Changing spend before the cause is clear turns a fluctuation into a conclusion. Record first, judge later is the basic discipline for these events.
Log the time of anomaly, the channels involved, the relative change in performance, and the external signs at that moment, marked 'cause to confirm'; check whether tracking, metric definitions, or platform rules changed, and write any attribution change into the changelog for before/after comparison. A judgement needs several periods of data — a single-point drop is not enough to characterise it.
This article covers recording and checking only; it issues no verdict on throttling and decides no one's channel switch. Record numbers during the anomaly as they are, write 'cause unconfirmed' when that is the case, without cross-market absolute comparisons; note flat numbers as flat and do not use them to push a direction change or a spend stop.