alpha is built from the full relative velocity, not its component perpendicular to the quarter-chord line

[1-Bort-1]

panel_inflow forms the polar-lookup angle from the full relative velocity:

v_eff = 0.5 .* (va_1 .+ va_2) .+ v_ind
alpha = atan(dot(v_eff, z_airf), dot(v_eff, x_airf))

(src/panel_aerodynamics.jl:119-120, and open-coded on the hot path at src/solver.jl:749-759.)

Gaunaa, Li & Pirrung, From thin airfoil theory to finite wings: consistent implementation of Lifting Line-based models, TORQUE 2026, J. Phys. Conf. Ser. 3224 042030 (OpenSourceAWE/Internal Papers/Gaunaa_2026.pdf) states the rule as CP1, boxed, p. 3:

Only the relative velocity component perpendicular to the local span direction, here defined as the quarter-chord bound vortex line, should be used in the inner 2D part of the LL framework.

and writes it as Eq. (1), p. 4, applied before the angle is formed:

V^⊥_{3D,i} = V{3D,i} − (V{3D,i} · e{s,i}) e{s,i}

α_{3/4,i} = atan2(V{inner,3/4,i} · e{n,i}, V{inner,3/4,i} · e{c,i}) (Eq. 3, p. 5)

The projection is missing here. It matters only in the denominator: z_airf is already perpendicular to the quarter-chord line, so dot(v, z_airf) == dot(v_perp, z_airf) exactly, but x_airf is the raw chord and carries a spanwise component on any swept panel.

Half of the rule is already obeyed elsewhere in the same function. The dynamic pressure is built from v_span = cross(v_eff, y_airf) (src/panel_aerodynamics.jl:121, v_a_dist at src/solver.jl:731-766), which is the span-perpendicular speed — simple sweep theory's cos²Λ about the quarter-chord line. So q is measured in the crossflow plane and alpha is not.

Size

On the swept, twisted panel test/panel/test_panel.jl already builds, where the chord sits 45.128° out of the span-perpendicular plane (cos(sweep) = 0.705526):

v_eff	alpha today	Eq. (1) then Eq. (3)
[1, 0, 0]	−14.299°	−27.411° … −20.097°
[1, 0.3, 0]	−13.473°	−35.422° … −26.647°
[1, 0, 0.1]	−20.003°	−36.679° … −27.721°
[1, 0.3, 0.1]	−19.205°	−46.540° … −36.668°

The range in the right-hand column is a genuine ambiguity in the paper, not measurement slop: it never defines e_c,i by equation, and the word "project" does not occur in it. Taking e_c,i as the raw chord unit vector gives the left figure; taking it as the chord re-normalised inside the span-perpendicular plane gives the right one, smaller by cos(sweep). Both are 13–27° from what the solver does now.

Over the reference wings at N=40, the largest angle between chord and the span-perpendicular plane is 15.86° (:rectangular), 0.00° (:curved), 0.00° (:elliptical) — and test/thesis_oriol_cayon.jl forms alpha the same way the solver does, so the suite pins whichever convention is chosen and cannot see this.

What I would do

Apply Eq. (1) to v_eff in panel_inflow before the atan, and take e_c,i as the chord re-normalised in that plane, so that alpha, q_dyn and z_airf are all measured about the same line. That needs deciding, because the paper leaves e_c,i open and because it moves every swept panel's polar lookup — it wants a swept, tapered validation case with a known answer, which the suite does not have. #273 deliberately left it alone.

This was raised for the record as item 1 of "Related, possibly by design" on #272, where it was pinned on y_airf rather than on x_airf; #272's item 2 there is already out of date, since q is the crossflow one. It gets its own issue because CP1 turns it from a question into a defect. Found while checking #273 against this paper.

[1-Bort-1]

Measuring how far CP1 moves alpha on the reference wings and the repo's kites, on top of #273's quarter-chord normal (it is what makes z_airf square to the span line). The e_c convention changes every swept panel's polar lookup and the thesis oracle, so I'll ask which one before changing panel_inflow and the solver's hot path. Leaving #318/#319 alone.

[1-Bort-1]

Question from VortexStepMethod.jl-317

Which chord should alpha be measured against on a swept panel?

A) Raw chord: project v perpendicular to the quarter-chord line (Eq. 1), then dot it with today's x_airf. The raw chord sits outside that plane, so the denominator is scaled by cos(sweep). The result is an angle in no plane.
B) Chord re-normalised in the span-perpendicular plane, as x_airf := y_airf × z_airf in panel_axes. On #273's frame that vector is exactly the in-plane chord (4.6e-15 over 1000 random panels). The panel frame becomes orthonormal, so every place that forms or inverts alpha follows at once: panel_inflow, the solver hot path, dir_inflow, alpha_geometric and correct_aoa. The cost is that the thesis oracle's pinned tangential moves on :rectangular (at most 15.9° out of plane), which is re-pinning the reference.

Measured on converged solves, CP1 alone on top of #273, max |Δalpha|:

• V3 CFD tip: B 1.3°, A 4.3°
• ram-air NeuralFoil fixture: B 3.0°, A 0.2°
• LEI ribs: B 0.3°, A 0.6°
• :rectangular: B 1.2°, A 0.6°
• curved, elliptical: 0°

The 13–27° in the issue holds only on #273's 45° test panel.

My recommendation: B, stacked on #273. The branch would change panel_axes plus the hot path, add a test on that panel, and re-pin the oracle's tangential. No swept validation case exists, so the PR would say it is unvalidated.