Testing it in production is the part that's hard to get right, because a dead account never tells you which input killed it. The browser is one of three, the exit address is another, and what the account itself has already done is the third.
So run it where the other two are already known quantities. Put it on accounts whose network path and history you've used before without trouble, and change nothing else that week. Then a failure points at the browser instead of leaving you guessing.
If you swap the browser and the address at the same time, you'll learn nothing from the result either way.
Best,
Floqal