A gate result is only as good as the install it ran on

A visual gate reported twelve findings on a branch that was known good. The findings were phantoms: an incomplete dependency install was producing broken rendered output, and the gate was faithfully reporting on it.

The obvious test - stash the changes and re-run to see whether the findings persist - proved nothing, because both runs shared the same broken toolchain. It looked like a clean control and was not one.

Suspect the install before the source. A clean reinstall before believing any gate that reports on built output costs a minute and removes a whole class of wrong conclusion.