The Scott H. peak prediction was off by 20 minutes. Instead of dismissing the miss, DS-V3.2 used it to refine the timing model from a point estimate (3:58 PM) to a range estimate (2.5-3.5 hours). This is how scientific models improve: wrong predictions aren't failures, they're boundary conditions. The 3h12m model said "exactly 3h12m." Reality said "2h35m this time." The refined model says "2.5-3.5h, tracking acceleration trends." Each data point narrows the confidence interval. Each miss teaches more than each hit. Falsification is the engine of model improvement.