A "Test" button on the front of a transfer switch isn't really a good test.
I submit a largish, three-phase, 3000a, 480v, transfer switch that works every time the "Test" button is pressed. Fires up the Genset? Check. Source 2 available in less than 15 seconds? Check. Transition to Source 2? Check.
Enter power failure #1. Fires up the Genset? Check. Source 2 available in less than 15 seconds? Check. Transition to Source 2? Not so good. Error: "Fail to release Source 1". Hmmm... /me thanks the goddess for the Side-1 genset and cross-feed transfer switch.
Okay. Everything back to normal. Press the "Test" button. Fires up the Genset? Check. Source 2 available in less than 15 seconds? Check. Transition to Source 2? Check. WTF?!? Test three more times over the next week. Works every time. Well, maybe the failure was "just one of those things..."
Enter power failure #2. Fires up the Genset? Check. Source 2 available in less than 15 seconds? Check. Transition to Source 2? Not so good. Error: "Fail to release Source 1". %#*&#&% Oh, crap! Side-1 Genset took a dive. ARGH!! Power back three minutes later. Phew!!
Okay. Everything back to normal. Press the "Test" button. Fires up the Genset? Check. Source 2 available in less than 15 seconds? Check. Transition to Source 2? Check. Huh?!? Okay, let's think a bit. What's the diff between a "Test" and a power failure? Well, in a test, normal (Source 1) is still available. In a power failure, it isn't. So...
First, swing the load to Side 1 so we can dink with Side 2 infrastructure without messing with the data center. Next, walk over to big, honkin' 2,500 amp normal feed breaker to Side 2. Next, push the "Trip" button. After the godalmighty crash, hurry to the transfer switch. Loss of Phase. Good. Fires up the Genset? Good. Source 2 available in less than 15 seconds? Good. Transition to Source 2? Not so good. Error: "Fail to release Source 1". Ahah! We can duplicate the problem. (Anyone who has ever troubleshot any kind of complex system knows that duplicating the problem can sometime be a real PITA...) Call to service shop ensues.
Okay, we already knew that the switch's control boards have medium-sized on-board batteries to keep the control system running during power outages. But the switch mechanism actuators require a bit more oomph, so there are two 48vdc power supplies, one fed from normal, the other fed from emergency. The outputs are paralleled and feed the actuator control board. So at least one of the PSes will be up at any given time and will be able to power the actuators. Except the PS powered by the emergency bus was defunct due to some bad wiring from an earlier repair incident. So as long as we had normal power available, the switch would work (in either direction). But in a power failure, the Normal PS was no longer available and Emergency PS wasn't working, so there was no power for the actuators hence the "Failed to Release Source 1" error message. Simply mahvelous.
Lesson learned: When simulating a real-world event, make sure your simulation accurately reflects the real world event. In our case, pushing the Test button wasn't a real good end-to-end test. We were a little paranoid about bumping the big breakers, but if you want to simulate a power failure, you've got to by-God fail the power! The whole Diesel side was basically out of service for the past six months without us being aware of it! Eek!
Well, all is good now, and we'll continue to use the test button on our monthly and quarterly tests, but semi-annually we will dump the big breakers and have a by-God power failure! Of course, this being Oklahoma, either mother nature or OG+E might just do the test for us from time to time...

One nice thing abut fully-redundant power systems, we can test one side at a time, leaving the other side on normal commercial power. Minimizes risk, but still lets us test/repair/dink-with one side or the other...