You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
gate-45: one comment in one stylesheet silences the gate for the WHOLE repo (#415 class) #421
Found by the all-65 survey of the #415/#420 comment class. Full table: fleet-board/findings/gates-comment-class.md.
The defect
The inline PYRM heredoc — the thing that looks like gate-45's checker — is correct. It masks comments before GUARD.search. Fixing it would change nothing.
The defect is in the global pre-pass at run-hydra-gates.sh ~L6862–6877, the _a11y_style_files | python3 -c block. It greps UNIVERSAL over open(p).read() — raw text.
Measured, one fixture, one variable:
css/motion.css with two real unguarded motion declarations, plus css/reset.css whose only universal reset is inside
/* TODO(a11y): we still need @media (prefers-reduced-motion: reduce) { *, *::before … } Not done yet — tracked in #999. */
→ [gate-45] prefers-reduced-motion: PASS
the same tree with that comment deleted → FAIL — 1 <style> block(s)/stylesheet(s) with motion
Why this one is worse than the rest of the class
The pre-pass sets HYDRA_RM_GLOBAL_GUARD=1, and the per-file checker then sys.exit(0)s on every file in the repo.
So this is not one missed finding. One comment, in one stylesheet, anywhere in the tree, silences gate-45 for the whole app — and the comment that does it is the one an author writes while acknowledging the a11y debt. Every [gate-45] PASS in the fleet is suspect until a rerun.
Shape of the fix
Route the pre-pass through the same comment handling the PYRM heredoc already uses (_mask_comments), so the two halves of the gate agree about what a comment is. Today the careful half is bypassed by the careless half.
⚠️ Needs a positive control first: the pre-pass must be shown to set the flag on a real guard before any arm asserting that it does not set it on a comment means anything.
⚠️ And a paired anti-widening arm: a genuine repo-wide @media (prefers-reduced-motion: reduce) block in a shared stylesheet must still set the flag. A mask that eats CSS rather than comments turns this gate from "silenceable" into "always red", which is the fix over-applied.
Fixes for the four most serious instances of this class (gates 7, 25, 48, 49) are in #425, which also carries the full 65-gate survey table and the method notes.
Activity
rubenvdlinde commentedon Aug 13, 2026
Fixes for the four most serious instances of this class (gates 7, 25, 48, 49) are in #425, which also carries the full 65-gate survey table and the method notes.
fix(gate-45): a commented-out universal reset silenced the gate for t…