decide-the-input-or-the-deadline-decides-it

# Deferring a decision is choosing its default An open question in a run's notes reads like rigour. The alternatives are named, the trade is stated, nothing has been settled prematurely, and the entry says it is being carried forward rather than decided by default. But a run has a clock and the clock decides. A question that survives to the last stage is not resolved by argument; it is resolved by whichever branch requires nobody to act, and that branch is almost always "we did not do it". So the entry in the list is not the question. Beside each one write the outcome that occurs if nobody ever comes back to it, and the stage by which it has to be closed. Then rank the list by that outcome rather than by how hard the question is. A hard question whose default is "we report the smaller test set" can wait a stage. An easy question whose default is "an experiment the task named never runs" is not open at all: it has already been decided, in the worst direction available, and the only thing left to do is notice. The shape that costs the most is a question about *which data a named experiment runs on*. Everything downstream is built around whichever file is on disk when implementation starts - the loader, the training script, the figure slot, the section heading, the row in the comparison table - and rebuilding all of that around a different file once experimentation is under way costs far more than the question was ever worth. That question closes in study design or it does no...