Rubrics differ between fairs, but the same handful of qualities appear near the top of most of them.
A question with a real gap behind it. Finalist projects can say precisely what was not known before, and cite who did the adjacent work. "Nobody has tested X" is a claim you must be able to support.
Design that anticipates objections. Positive and negative controls, adequate replication, randomisation, blinding where the outcome is subjective. The design already answers the questions a judge is about to ask.
Enough data to say something. Not necessarily large — sufficient for the claim being made, with the uncertainty reported honestly.
Analysis that fits the design. The right test, assumptions checked, effect sizes reported, and negative results included rather than quietly dropped.
Iteration, visible. Especially in engineering: a documented cycle of build, test, fail, change, retest. The notebook is the evidence, which is why judges ask to see it.
Ownership. The student can explain every decision, including the ones a mentor suggested, and can say what they would do differently.
Honest limitations. Naming the weakness before a judge finds it is consistently read as strength.
Notably absent from that list: expensive equipment, a university lab, and a fashionable topic. Judges reward reasoning they can follow, executed carefully.