If you're alerting when alert conditions begin / end, then (for this test) you're going to be controlling alert frequency by increasing samples to examine. I'd probably set this to 0 minutes, and also set the minutes to wait to 0 minutes as well. Then, set the "Samples to examine" to a number that's big enough to control over-alerting from up/down "flapping".

If you set the samples to examine to 4320, it does mean that you're looking at the most recent 4320 samples when evaluating the alert. An outage will stay "in scope" of these 4320 samples until you've collected 4320 good samples, so the alert won't go back to "everything is ok" condition for quite some time. This controls the flapping.

Let us know how it goes.