A 1:5 Froude-scaled model of a cargo aircraft flies at a Reynolds number where the tail's lift slope is poorly known. In the aerolog-scaling programme, with the true slope 0.75 to 1.0 of the surrogate's, none of 78 mode-matched tails kept the gust response within 10% at every slope. A tail-area multiplier set in flight did: 67 of 78 designs stayed within 10% on every gust peak.
How well the model must be identified. The fix is only as good as the in-flight estimate of the short-period mode. A fit with 2% error leaves 84 to 86% of noisy draws within 10%; at 5% error it is 58 to 61%, and no design is reliable. A least-squares fit of a noisy free response gets worse with longer records (0.33%, 0.70%, 4.45% frequency error at 1, 2 and 4 phugoid periods); an output-error fit does not (0.21% to 0.33%).
Limits. Simulation only, with an unmeasured tail slope, white sensor noise, no turbulence and a known initial state. The paper draft now has this as Section 4.10; the project log has every run and prediction.