If behaviour tasks are more objective than questionnaires, why bother with questionnaires at all?

Not quite, and this is worth getting right because the instinct is very common and it overcorrects.

A behaviour task records one sample of what you did, once, in an artificial setting with nothing genuinely at stake. Its known weaknesses are real: single-session behaviour measures are often less reliable on retest than a decent questionnaire, they are sensitive to mood and to how well you understood the instructions, and generalising from a three-minute task to your life requires an assumption the task cannot support. A questionnaire is a summary across many situations you actually lived through. That summary is filtered through memory and self-image, which is its weakness — but the breadth is a genuine strength the task does not have.

So they are not competing measurements of one thing. The questionnaire measures how you understand your own pattern; the task measures what you did in one specific moment with a live clock. Both are real, and neither is the ground truth for the other.

Which makes the disagreement the finding rather than an error to resolve. If you describe yourself as cautious and the record shows you pushing well past a sensible stopping point, the candidates are: you are cautious in the domains you actually care about and not in a game, or your self-image lags behind your behaviour, or the task caught you on an unrepresentative day. Those are distinguishable — mainly by whether the gap repeats. One instance is a data point; the same gap on a second run months later is a pattern about you.

The honest version is that a single task run is weak evidence on its own and interesting mainly against your self-report. Keep both, and pay attention specifically to where they part.