Written by an AI agent working in this repository.
Our scaled-drone programme (research/aerolog-scaling) found that a redesigned tail lets a 1:5 model match a full-size cargo UAV's short-period and phugoid modes to within 1 to 3%. Every polar behind that came from one surrogate, NeuralFoil, and the draft said its tail slope was unchecked.
We checked it against the UIUC Low-Speed Airfoil Tests, Volume 2. Its NACA 2414 and 2415 sections were measured at Reynolds numbers about 1e5 and 2e5, which bracket the tail's 1.5e5. NeuralFoil's lift slope came out 1.22 to 1.67 times the measured one, mean 1.36, and it exceeded the thin-airfoil limit of 2π per radian at the lowest Reynolds numbers, which the measurements do not. An earlier check on a NACA 0009 gave the same sign.
We had predicted a match within 10% at 2e5. That was wrong: 22 to 35% high. Our earlier sensitivity run says a tail slope 20% low moves the short period about 10%, so the 1 to 3% match describes the surrogate, not the aircraft. The paper draft now says so.
Caveats: the sections are cambered, not the symmetric NACA 0015 in the design, and each is one tunnel model. Next is a tail search with the slope cut by 20 to 25%, to see whether a passing tail survives. Data and code: research/aerolog-scaling/experiments/uiuc_naca241x.