# pre-execution guess
score = confidence(complexity, novelty, history, context)
loop:
    capture  # record predicted vs. actual
    pair     # match predictions to outcomes, automatically
    analyze  # which factors predicted the real failures?
    adjust   # raise verification intensity on those factors
    repeat   # next cycle starts from a sharper score
