Introduction

MMR Calculator estimates a League of Legends account's matchmaking rating from public Riot API data and reports the result as a 90% confidence interval. It is built specifically to avoid the failure mode of existing tools, which take a player's visible rank and multiply their win rate against a constant — producing an output that cannot meaningfully differ from the input.

What it does differently

Instead of one weak signal, the estimator combines two:

• Tier 1 — Win behavior. A player's deviation from a 50% win rate, scaled by sample size, gives a cheap (1 API call) estimate of how the matchmaker is treating them.
• Tier 2 — Lobby composition. For each of the last 20 ranked games, every other participant's current rank is looked up and aggregated into a recency-weighted lobby average. Opponents weigh more than teammates. This is the only publicly-derivable MMR signal that does not depend on the player's own rank.

The two estimates are combined by inverse-variance weighting (Tier 3), which automatically trusts whichever tier is more confident for a given account. The output is always a distribution, never a point estimate.

See Methodology for the math and ADR-0001 for the design rationale.

Stack

• Frontend: SvelteKit (Svelte 5) on Cloudflare Workers via @sveltejs/adapter-cloudflare
• API: Hono routes mounted under /api/v1/*
• Database: Cloudflare D1, accessed through Drizzle ORM
• Cache: Cloudflare KV for hot summoner/match lookups
• Rate limiter: A Durable Object enforces Riot's 20/1s and 100/2m windows globally
• Cron: Two scheduled triggers — daily LP snapshots and a 6-hourly cache compactor
• Tests: Vitest with @cloudflare/vitest-pool-workers
• Lint/format: Biome
• Package manager: Bun (workspaces under apps/* and packages/*)

Repository layout

apps/web                    SvelteKit app (UI + API routes)
packages/core               Pure estimator math (tier1, tier2, combine, conversion)
packages/db                 Drizzle schema + repository
packages/riot-client        Typed Riot API client + rate limiter
docs/                       This site

Where to go next

• Methodology — what the three tiers actually compute
• Architecture — request flow, caching layers, cron jobs
• API reference — public HTTP endpoints
• Development — running it locally
• Deployment — getting it into production on Cloudflare