[ashtonanderson]

Fourth round, after #324: the page-loading costs Lighthouse found on maiachess.com (phone scores 41–70). Details and all numbers in LOADING.md ("Round 4").

Changes

• Engines only where they're needed (8383bc0).
  • The /broadcast list started Stockfish and Maia: _app.tsx matched any path containing /broadcast, so a first visit downloaded Stockfish's 76 MB of nets just to show a list. It now starts no engine; a round starts them when it opens.
  • On /drills and /puzzles, which need Stockfish only after a while, it now starts once the page has loaded and the browser is idle (deferStart).
• Engine code only on engine pages (017694c). _app.tsx imported the Stockfish, Maia and analysis-list providers, so every page downloaded and ran both engines' code and Maia's move tables (111 KB, parsed on every load). Engine pages now list the providers they need (Page.providers, src/contexts/pageProviders.ts). First Load JS shared by all pages 308 → 235 kB.
• Headings at once (1ab0ce1, 8fb8652).
  • The leaderboard and the broadcast list rendered nothing, on the server too, until their data arrived.
  • The leaderboard now renders its heading and columns at once, with placeholder rows until the rankings arrive.
  • The broadcast list renders its heading while Lichess answers, and no longer waits for the visitor's session (it asks only Lichess).
• Home page counters (181412c, d6ef8b9): a fixed count-up instead of a spring that took ~3 s to settle, re-laying out the hero on every frame; 1.5 s in 181412c, shortened to 0.5 s in d6ef8b9.

Results (main → branch)

Emulated phone (4G, every thread at a quarter speed), medians of 5 loads:

/broadcast list, first visit	no 76 MB download; main-thread work 9.2 → 4.3 s; main content 4.8 → 0.7 s
/leaderboard	main content 3.2 → 0.9 s (first visit), 1.8 → 0.8 s (returning); main-thread work unchanged
Home, first visit	main-thread work 11.0 → 7.9 s; JavaScript 636 → 474 KB
/turing, /play, returning	main-thread work 2.8 → 2.2 s, 1.9 → 1.5 s
/drills, /puzzles, returning	Stockfish starts 0.8–1.6 s later; main-thread work 5.0 → 3.9 s, 6.0 → 5.7 s

Home on a desktop: main-thread work over the load and 6 s 2.65 → 2.04 s (counters at 1.5 s).

The home page's other costs, measured by leaving each out (phone, main-thread work): without the Play section's 36-board grid 6.8 → 5.8 s on a first visit, 5.1 → 4.3 s returning (desktop 1.29 → 0.94 s); without the game carousel, no measurable change. Neither is removed here.

Lighthouse, phone settings, local builds, medians of 3:

page	score
home	58 → 67
/broadcast	43 → 71
/leaderboard	52 → 73
/turing	51 → 64
/play	72 → 76
/blog, /drills, /puzzles	no change

Stockfish

Not slower (scripts/perf/chart-probe.mjs, 8 moves per run, analysis page):

	main → branch
Desktop: depth 12 after a move	0.04 ± 0.05 s sooner (16 of 32 moves)
Desktop: depth reached before the next move	16.66 → 16.88
Phone: depth 12 after a move	0.06 ± 0.22 s sooner (27 of 47 moves)
Phone: moves reaching depth 12	48 → 47 of 48
Phone: depth reached before the next move	14.62 → 14.54

Testing

• npm test (84, with main's), npx tsc --noEmit and npm run lint pass, rebased on main after Capture PostHog web vitals again #330 and Track what visitors do, and stop losing events #332.
• Checked in local production builds:
  • Analysis gets Stockfish's evaluation, loaded directly and after an in-app move from the home page (0.76–1.0 s on both builds).
  • Puzzles show Stockfish's evaluation as soon as a puzzle is given up.
  • A drill started right away loads as on main.
  • The broadcast list requests no engine file, and opening a round from it starts Stockfish.
  • The leaderboard's placeholder rows line up with the loaded rows (layout shift 0.0001).
• Not tried on real phones or in Safari.

Not done

Shells for the app pages. /puzzles, /drills, Bot or Not and analysis still send only the header and footer. They wait for the visitor's session and their data, and choose the phone or desktop layout in JavaScript after loading. A server-rendered shell would show the desktop layout on a phone and then jump, so it first needs that switch moved to CSS.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
maia-platform-frontend	Ready	Preview	Nov 3, 2026 6:32am UTC