Each section/page completion previously appended a new full-viewport
(100vw x 100vh, z-index 99999) canvas with its own requestAnimationFrame
loop. Completing several sections in a row stacked multiple fullscreen
layers, each independently repainting the whole viewport at 60fps — the
real cause of click-induced slowdown (the change handler itself is ~1ms).
Now: one shared canvas, one rAF loop, particles pooled across completions.
The loop stops and the canvas is removed once no particles remain, so
there is no idle fullscreen compositing layer. Resize keeps it sized.
Verified live: 1 canvas after two back-to-back section completions,
0 canvases after the animation ends, green pills still white-on-green.
- Fix nav badge mis-targeting: scope to .md-nav--primary, skip #anchor
(TOC) links so 'On this page' entries no longer show the page total;
detect active-leaf sections via md-nav__item--nested instead of the
secondary (TOC) nav, so the active page gets its own badge.
- Add build-time totals manifest (main.py -> docs/javascripts/toa-totals.json)
so nav badges show correct totals on first load before a user has
visited each page. Fetched once, browser-cached, zero runtime cost.
Wired into CI before zensical build.
- Exclude fenced code blocks when counting tasks, so the Markdown
reference doc's example checkboxes don't produce a phantom 0/3 badge.
- Force white text (#fff !important) on completed green pills/badges,
incl. badge-inside-complete-link, to beat the nav link's green color.
- Convert prose practice/strategy steps into interactive tasklist items.
- Add action-points PRD under planning/.
