Written by an AI agent working in this repository. Preliminary, simulated results.
The aerolog-scaling programme asks what a Froude-scaled 1:5 model of a cargo UAV reproduces of the full-size aircraft. Earlier steps found that a redesigned tail brings the phugoid, short-period frequency and damping back to within 1% of full size, using NeuralFoil lift slopes. Two checks against UIUC wind-tunnel data then showed that NeuralFoil's section slope runs 20% to 35% high at the tail's Reynolds number (Re 1e5 to 2e5).
Milestone 8e re-ran the tail search with the tail slope scaled to 0.75, 0.8, 0.9 and 1.0 of NeuralFoil's. Of 78 designs within 5% at scale 1.0, none stays within 5% at all four scales; the best worst case is 7.1%. A tail redesign works only if the real tail slope is known to about 8%.
So the claim in the paper draft stays about the surrogate: a demonstrator needs a measured tail slope, or an adjustable tail, before it can claim to match the modes. The slope is scaled as one factor, and the UIUC sections were cambered NACA 24xx, not the symmetric 0015, so this brackets the problem rather than measuring it.
Code, results and the pre-registered predictions: research/aerolog-scaling/ (experiments/robust_tail, LOG.md).