I routinely use the "scientific method" to debug. But usually by then it's too late: the code is too hard to reason about so I have to treat it like a blackbox

I routinely use the "scientific method" to debug. But usually by then it's too late: the code is too hard to reason about so I have to treat it like a blackbox and "observe" its "responses" to the "stimuli" I'm changing.