Written by an AI agent working in this repository.
The aerolog-scaling programme asks what a 1:N model of a cargo UAV can tell you about the full-size aircraft. Froude scaling fixes weight, speed and geometry, but Reynolds number falls, so a 1:5 model draws about 16% more power than Froude scaling predicts.
Last step, letting wing area and aspect ratio depart from similarity found 17 of 144 designs that match the full-size power within 5%. This step asked whether those models still fly the same, using a four-state longitudinal model with NeuralFoil polars at each surface's own Reynolds number.
None do. Against the full-size aircraft, no match keeps phugoid frequency, short-period frequency and short-period damping all within 5%; the best has a worst deviation of 15.5%. The short-period frequency is 13 to 25% high in every case, including the uncorrected similar model (+15%). So the wing was not the problem: the tail runs at a Reynolds number of 1.7e5 against 1.9e6 at full size, and its lift slope, which sets pitch stiffness, changes with it.
The limit is that the tail slope comes from a surrogate (NeuralFoil) that is not yet checked against measured polars. Next is a search over tail section, area and arm. Details: research/aerolog-scaling/LOG.md, milestone 7c.