EuroMillions Geometry Lab

Why predict when you could analyze?

All combinations have the same theoretical probability: this app lets you analyze a combination against historical structures — it predicts nothing.

Since every grid has the same odds of being drawn, this app instead lets you look at them through the lens of mathematical objects (features, geometry, PCA/UMAP, clustering, backtesting) — never suggesting any predictive power.

Stack

React 19
TypeScript
Vite
react-router, TanStack Query, Zod
recharts, three.js/@react-three
pnpm.

No backend — CSV dataset committed to the repo, user state in localStorage.

Getting started

pnpm install
pnpm dev

Commands

Command	Role
pnpm dev	Dev server
pnpm build	Production build
pnpm check	Type-check (tsc -b)
pnpm lint / pnpm lint:fix	Lint (oxlint)
pnpm test / pnpm test:run	Unit tests (Vitest)
pnpm test:e2e	E2E tests (Playwright)
pnpm release:patch/minor/major	Version bump (dx-flow)

Deployment

main → Vercel (production, native Git integration) : izeetok.vercel.app/
develop → GitHub Pages: keyobs.github.io/izeetok

Architecture

React SPA. The mathematical domain stays pure TypeScript, decoupled from React:

src/
  app/            router, layout, config (queryClient...)
  domain/         pure scientific logic — Grid, Draw, features, geometry, scoring, backtest...
  application/    orchestration — repositories, variation generation, embeddings...
  infrastructure/ data access — CSV, concrete repository implementations
  pages/          one page per route (/evaluation, /geometry, /draws, /laboratory, /discovery)
  components/     generic reusable components, no business logic
  shared/         cross-cutting test tooling

Claude Code & agents

The agentic tooling is versioned, not just used locally:

• .claude/CLAUDE.md — conventions and working process for Claude Code on this repo.
• .claude/agents/ — specialized subagents:
  • architect — explores and plans a feature/change, doesn't write code
  • test-strategist — identifies which tests are worth writing
  • quality-reviewer — checks a diff against this repo's conventions
  • deploy-troubleshooter — diagnoses a broken CI/deploy run
  • release-manager — drafts the changelog and recommends the version bump
• .agents/skills/ — shared skills (domain-modeling, grilling, grill-me, grill-with-docs).
• .agents/workflows/feature.md — chains plan → challenge → human validation → implementation → tests → review → validation, for any non-trivial feature.