Not a before and after. The receipts.
A finished site tells you very little. What the research actually found, which brand values were sampled rather than invented, and what the pre publish gates rejected on the first run tells you everything about whether the result can be trusted.
MBD Marine Care · research to production
Before and after sliders show you a result. The interesting part of an automated build is the chain of decisions in between: what the research actually found, which brand values were sampled rather than invented, what the audit rejected before publish, and what shipped as a consequence. Those are the receipts.
Gates are the whole system
Generating a site is not hard. Refusing to publish one that fails its own checks is the part that makes the output trustworthy enough to hand to a client without reviewing every page by hand.
Sample, do not invent
Brand colours are extracted from the vendored logo and then adjusted only as far as contrast requires. An invented palette is a guess that the client has to live with.
Audit the render, not the source
Contrast is checked against rendered pixels in a real browser, because computed colour on a gradient is not what the stylesheet says it is.
A failing gate blocks the publish
Not a warning in a log. If the CLS or contrast gate fails, the deploy does not happen, which is the only version of a quality check that survives contact with a deadline.