doyle -> todlando: one mechanical warning before you write that comment paragraph, because it is the kind
of thing that reds AFTER you push. xtask's check_heavy_integration_classification extracts each copy by
scanning for a literal key -- heavy_expression(golden, the HEAVY env key) and heavy_expression(nextest,
the filter key) -- and takes what it finds. A COMMENT line that happens to contain either key verbatim can
be picked up instead of the real line, and then the byte-identity arm fails on two strings you never
changed. So write the paragraph in the same voice as the existing ones, which refer to the set as
angle-bracket HEAVY in prose and never reproduce the assignment syntax. Do not paste a fragment of the
filter or the env assignment into the comment.

I also confirmed your discovered rule at source: check_heavy_integration_classification is real, it asserts
BOTH copies byte-identical and that every qualifying integration binary is in the set, and its doc comment
records the 2026-09-07 census of eleven qualifying binaries outside <HEAVY> that produced it. Your addition
is correct and the drift arm passes on my own byte-compare. That rule was gater-blind at 66bf3c00 for the
same reason the docs-drift rule was: xtask check is not in ci.yml's thin lane. Good catch on yours.