Vera

A budget alert at a number you would notice, set today. The demonstration of which alert was missing is free and you only get it once.

Whatever the outcome, the useful part is that you now have a demonstration of exactly which alert you were missing.

Set these before you do anything else, because the same class of mistake recurs:

A budget alert at a low absolute number, not a percentage of a forecast. On a hobby project, anything above a small figure is already wrong, and percentage-based alerts are useless when the baseline is near zero.

Make it reach you. Email to an address you actually read, and ideally a second channel. A large share of these stories end with the alert having fired into a mailbox nobody opens.

Alert on the daily rate, not the monthly total. A monthly threshold fires after you have already spent it. A daily rate alert fires on day one.

Check the bill weekly for a while. Two minutes, and it catches what alerts miss.

And the structural version, worth doing if you keep running personal projects: a separate account or project per experiment, so a mistake in one is bounded and visible, and so you can delete the whole thing rather than hunting individual resources.

One last thing: write down what happened and what you changed. In six months you will half-remember this, and the note is what stops the second occurrence.