typrlab

Live → typrlab.com

An adaptive typing trainer that figures out exactly what's slowing you down — key by key, transition by transition — and drills it, so practice time goes where it actually moves the needle.

For anyone who types a lot — students, writers, content creators, developers, and other professionals — not just programmers. That's why typrlab ships polished light and dark themes (research shows the best choice depends on your environment and eyes), including a warm sepia Paper theme for long writing sessions.

typrlab generates practice text on the fly from the letters you've unlocked, weighted toward your weak spots. It expands your alphabet only when you're both fast and accurate, tracks per-key and per-transition timing, and shows you an honest, actionable picture of what to work on next.

Status

🟢 Runnable. Next.js 14 (App Router) + TypeScript, local-first (your data stays in your browser). Real, deep-linkable routes: /, /analysis, /settings.

npm install
npm run dev          # http://localhost:3000 — just start typing
npm run test:smoke   # headless checks of the adaptive core
npm run typecheck    # tsc (strict)
npm run lint         # ESLint
npm run build        # production build

Features

Adaptive practice

• Guided lessons over a dynamically expanding letter set (starts with a small core set and grows as you improve).
• Configurable key-introduction order — Balanced (default; frequency-driven but alternated across hands), Frequency, or Home row (the classic touch-typing path). A learning-path preference, not a speed claim; there's no proven-optimal order (rationale in src/core/keyOrder.ts).
• Stop-on-error typing with precise per-keystroke timing and outlier rejection.
• Accuracy-aware unlocking — a key must be typed fast and accurately before it counts toward unlocking the next letter.
• Bigram-aware targeting — finds the slowest transitions (digraphs like th, er) hiding inside otherwise-fine per-key averages, and drills them.
• Phonetic pseudo-words plus a real-word mode.
• Calm, configurable typing cursor — Box / Underline / Line / Block — and a theme-aware confidence ramp (each theme's own red→green, mixed in OKLCH).
• Adjustable text size (the evidence-backed lever for reading performance) and a typing-font picker — 9 self-hosted fonts (no CDN), three per category: monospace (JetBrains Mono, Cascadia Code, Source Code Pro), sans (default — Atkinson Hyperlegible, an accessibility-first legible face), and serif. Typeface is largely preference — research shows it barely affects reading speed; size and personal fit matter more (see docs/font-research.md).

Content modes

• Practice on what you want: Adaptive (the guided stream), Words (real words, ungated), Numbers (configurable digit groups), or your own pasted Custom text. Optional capitals and punctuation modifiers sprinkle into the adaptive and words streams.
• Practice layouts — Coach (a focus rail showing what's being drilled and why), Instrument (a dense stat strip), plus a chrome-free Zen focus mode (Esc to exit). Per-key stats accumulate across every mode.
• Lesson length — choose how many words per line (5 / 10 / 15 / 25).

Everything is tunable in Settings — practice mode, lesson length, a daily practice goal, key-introduction order, plus theme and font pickers:

A progress dashboard answering three questions at a glance — am I improving?, what's still slow?, have I been consistent?

• Scorecards — best speed, avg of your last 10 (with trend), accuracy, time invested, day streak, and letters unlocked.
• Speed over time — net + raw wpm with a goal line; plus accuracy and a consistency (rolling std-dev) trend.
• Per-key speed heatmap — your QWERTY keys colored by current speed (with per-key wpm); locked keys dim.
• Slowest keys (avg ms vs. target), accuracy by key (lowest first), and slowest transitions (the digraphs costing you the most).
• Practice calendar — a GitHub-style activity grid with your current/best day streak.

Themes

• 10 presets — 6 dark + 4 light — chosen for a broad audience: dark dev favorites (Dracula, Tokyo Night, Nord, Catppuccin Mocha, One Dark) and reading/writing-friendly lights (a warm Paper sepia, Solarized Light, GitHub Light, Catppuccin Latte), plus typrlab's default Amber.
• Palettes are defined as OKLCH design tokens (perceptually uniform), generated from each project's canonical colors by scripts/gen-themes.mjs.
• Your choice persists locally.

Your data

• Local-first. Full JSON export / import of your history and settings.

Tech & structure

Next.js (App Router) · TypeScript (strict) · ESLint + Prettier · self-hosted fonts (no CDN). Ships as a fully static export (output: 'export') — the app is client-side and local-first (ssr: false), so it hosts on any CDN (see DEPLOY.md). Server features later (SEO pages, sync/payments) would drop the export flag and add sibling routes — no app rewrite.

• app/ — routes (/, /analysis, /settings) + shell (layout, no-flash theme script)
• src/core/ — framework-agnostic domain: typing engine, per-key & per-transition stats, phonetic model, adaptive lesson generation
• src/ui/ — React components (board, keyboard, charts, settings)
• src/hooks/ — useTypingSession (all stateful logic, kept out of the view)
• scripts/ — generators (gen-themes, gen-words) + a headless smoke test

Roadmap

• More content modes: code (with an on-screen number/symbol keyboard) and quotes (a public-domain prose corpus).
• A custom theme builder; optional auto light/dark by time of day.
• Larger / real phonetic models per language; responsive / mobile polish.
• Optional accounts + cross-device sync; SEO / marketing pages.
• CI (typecheck / lint / build / smoke) once the workflow OAuth scope is granted.

Acknowledgements

typrlab's adaptive approach was informed by studying keybr.com (source, AGPL-3.0) — an excellent open-source typing trainer. typrlab is an independent, from-scratch implementation with its own design and improvements; no keybr code was used. Detailed study notes: docs/keybr-explainer.md. See docs/ATTRIBUTION.md.

License

Proprietary — all rights reserved. See LICENSE.