[hliyan]

I do believe that "do something for a set period and see what happens" is the wrong way to look at timeboxing.

Rather, (at least to me) it's "don't extend the deadline, instead reduce the scope". You slice your deliverables small enough so that if you miss the deadline, it's not an all-or-nothing situation. You can always deliver a usable subset of what you planned/forecast.

Take your example, "writing those two missing features today". You might just end up delivering one of those features if your forecast turns out to be inaccurate. It's more effective if the timebox is larger and the tasks are smaller.