Eval fixtures Shipped
The intent contract and scenarios live inside the team directory, so an installed copy carries the thing it is judged against.
An intent contract declares the mission, the command surface, and at least three weighted excellence criteria — at least three because a coarser rubric cannot steer a loop. A scenario is the exact argument list to run, with a single supported placeholder for the output directory. Fixtures are hashed, so a run records precisely which version of the rubric its scores were judged against; scores across a fixture edit are not comparable and the system knows it.
Why it lives inside the team
A team you can install is a team someone else can run — and a score means nothing without the rubric it was measured against. Shipping the fixture with the team is what makes "this team scores 87 on its own criteria" a checkable statement rather than a marketing one.