376 pages, 376 tests, 0 failed - all ok
Every test here is a page of the manual application,
fetched over HTTP a dozen at a time, and compared with the prediction of the same
name in apps/regression/manual/. The pages stay in their own application;
the store holds nothing but what each must answer. An index is fetched full, everything
else bare with &padInclude, the same rule the harvest applies - and a page
that draws answers a /pattern/ pinning its skeleton, everything else an
exact body.