/ui/led-ring-mirror card drift after kid/smart toggle #30

(observed 2026-04-28; scope tightened 2026-04-29)

The toggle-button leg of this drift was diagnosed and fixed in dotty-stackchan 3646e93 (dashboard: read kid_mode fresh + add 10s poll to toggle cards) — kid_mode getter was a closure over a stale module-level KID_MODE global, and neither toggle card had a time-based refresh trigger.

This issue now tracks only the separate LED-ring-mirror card at /ui/led-ring-mirror, which reads from _perception_state (SSE-cached) and can still disagree with the physical pixels at 8 (kid) / 9 (smart) after a toggle.

Reproduce + capture which leg lags

• (a) Flip via dashboard — does the physical pip update within ~1 s? Does the mirror dot update?
• (b) Flip via voice — same two checks.
• (c) Flip via /admin/kid-mode curl — same.

Suspects to triage

Post-2026-04-29 hot-load, the dashboard path no longer restarts the bridge, so the "mirror polls a dead bridge" hypothesis is out. Remaining suspects:

• SSE dotty-refresh events may not fire on toggle-source paths.
• Firmware state-file read on boot vs. live MCP push race.

Once captured, fix is likely small (single missing event emission or missing live-state push); the value here is precise repro before guessing.

Migrated from local tasks.md on 2026-05-17.