• Absolute scales are not settled. Our α disagrees with the published α by 1.03× to 2.08×, and not by a constant — a constant would be a convention and could be absorbed. Every TeV and GeV here inherits that. The mass ratio, the bill in eighths and the two arithmetic laws do not: no normalisation enters them.
• One published case is the whole validation of the calculator — AHMN's 1.2046. The engine reproduces it to 1.2045. Nothing else anchors it.
• Three anchored models, and one builder that is not anchored to anything. The three — SU(7) on S¹/Z₂ × S¹/Z₂, SU(4) on T²/Z₂ and Haba–Yamashita's SU(3) on S¹/Z₂ — have fixed orbifolds, fixed conventions and a fixed catalogue; what you vary in them is multiplicities. The SU(N) builder is the other thing: there the model is the input, any SU(N) and any boundary condition on S¹/Z₂, and no data file in this tool has heard of it. What it does not have: one orbifold only (S¹/Z₂ — T²/Z₃ and T²/Z₆ are counted and classified here but carry no potential), no brane matter even though the anomaly ledger says brane fermions pay the bill, no Yukawa couplings anywhere, and the vacuum is located for one and two Wilson-line phases and declines past that, which covers every boundary condition of SU(N) up to N = 5, 95% of them at N = 6 and 62% at N = 10.
• One loop, no running, no matching. There is no RGE evolution to the weak scale, no threshold matching, no scheme dependence and no error bars. The Higgs mass here is the curvature of a one-loop Wilson-line potential, not a prediction you can confront with data.
• It is a page, not a library. There is no Python API and no LHE/SLHA/UFO export. What leaves is the result card as JSON and as plain text, and the model as LaTeX with its BibTeX — every value carrying its own honesty label as a field. If you need this inside a pipeline, take the card and the formulas, not the page.

What it is good for: screening a candidate content on the exact, normalisation-free quantities before anyone computes anything; checking your own implementation against the one anchored case; and seeing which claims of Parts III–VII survive a computation that does not share their code.