Campaign performance is checked when somebody remembers, which is usually after it has been bad for a fortnight. A daily read makes the decision small and routine.
Why the criteria are forgiving
A monitor that kills campaigns on thin evidence is worse than no monitor, because it destroys the volume that would have produced the answer. Daily numbers on a single campaign are noisy enough that a strict rule will kill a good campaign roughly as often as a bad one.
So the bar for a kill is high: genuinely dead, or underperforming consistently over a real window. Stay is the default and is the correct answer most days for most campaigns.
Reading splits separately
A campaign sending across several pools is several experiments wearing one name. When one pool has a deliverability problem its numbers drag the campaign average down, and the campaign looks like a copy failure. Evaluating each split on its own separates the two, and the distinction usually changes the verdict.
It only reports
The monitor never pauses, scales or edits anything. It produces a verdict and posts it. Acting is a separate skill with its own confirmation step, because a monitor that also acts is one that can quietly make a bad call at seven in the morning with nobody watching.
Runs well with
Skills compound. These are the ones we usually install alongside it.