Five things that are new.
The output is correct by design when the design system makes the wrong build impossible, not discouraged.
A design system built to the thesis.
Not a slide about the thesis. A complete design system that holds to it: tokens defined once, components where the wrong build will not compile, documentation that ships with the code.
Then a coding agent built a working application from it, with no design direction. The result is the system's output rather than a designer's, which is the whole claim made concrete.
Six properties, each scored with the evidence attached.
Any design system can be assessed against the thesis, by anyone, in two passes.
Where something can't be measured, it comes back as not measured, with the reason. That never renders as a failure.
Ten systems, read and built with.
The same instrument, the same frozen task, applied to ten well-known design systems: what each one holds, and what an agent actually does with it.
Where a property can't be read, it comes back as not measured, with the reason, never as a failure. The protocol was written and frozen in July 2026, before any run happened, and any change since is published as a deviation with its date and reason.
Use it.
The definition, the six properties and the diagrams are published under CC BY 4.0. Cite them, build to them, argue with them.