Everything you ever wanted to know about how compatibility is computed.
What is AniMatch?
AniMatch is an unofficial companion tool for AniList that compares two user profiles and assigns a compatibility score. The score answers a simple question: “How much do our anime tastes overlap?” Every number shown on the match page is backed by public AniList profile data and the heuristics documented below.
The production site lives at animatch.raiku.dev. All processing happens server-side and the rendered page simply explains how the result was produced.
Where does the data come from?
- AniList GraphQL API: Profile statistics, favourites, detailed genre/tag breakdowns, release years, studios, staff and media IDs are fetched via AniList’s public API. No scraping is involved.
- User-supplied avatars: Avatar URLs are provided by AniList and cached only to draw the share image.
- Media art: Cover art, character images, and voice actor portraits originate from AniList responses and are hot-linked; no third-party CDNs are queried.
- Analytics: Lightweight request counts are logged so the developer can monitor uptime. No personal data or cookies are stored.
All cached assets stay inside /data on the host and can be invalidated without contacting AniList again.
How is the score calculated?
- Both usernames are validated and their AniList profiles are requested in a single GraphQL query.
- The response is split into thematic buckets (watch stats, genres, tags, favourites, etc.).
- Each bucket is handed to a heuristic from
scoring.jsthat returns a normalized score (0–1) and a weighting factor. - The weighted average of all heuristic scores is mapped onto a fixed spread measured across real pairs of AniList users, so each rank covers the same share of pairs: roughly the top tenth of pairs reach S, and the top half-percent SSS.
- The weights themselves were fitted to real profiles: each was tuned to predict how two people rate titles the score never looked at.
- The server stores the finished percentage plus avatars to generate a shareable OpenGraph image.
None of the heuristics rely on hidden data. If something shows up on AniMatch it can be traced back to the underlying AniList numbers.
Statistics, Purpose & Weighting
The table below enumerates every heuristic that feeds the final score. Each one is designed to answer a specific “Do we like the same things?” question. Weights are tuned so no single metric overpowers the others.
| Statistic | What it measures | Why it exists | Weight rationale |
|---|
Compatibility Ranks
Every score is mapped to a rank. The list below shows each bracket, its badge, and what the range represents.
Caching & Images
- Once a comparison finishes, the server stores the two usernames, their avatar links, the percentage and the date it was worked out, so the share image can always be drawn again.
- No personal AniList tokens are saved. Only publicly accessible data makes it into the cache.
- The generated OpenGraph images referenced on Twitter/Discord cards are served from
/share; each one only ever shows the latest score for that pair.
Need More Detail?
All business logic lives in matchmaker.js, while the Express server proxies AniList calls and hosts the static site.
Questions or tuning suggestions? Reach out to raikutro@raiku.dev or @raikutro on Discord.
What AniMatch Keeps
A match itself is worked out in your browser and stored nowhere. Separately, so that suggested matches can rank people without re-reading every profile, the server keeps a small index: for every username it has been asked to match — in either box — and every username searched for on that page, it reads the public profile once and stores a 512-byte signature of it, refreshed at most weekly.
The whole database record stored is a username and its signature/hash. Most of the signature is collision-laden bitmaps plus a few rounded figures (how you rate each genre and stretch of release years against your own average, and roughly how much you have watched), which can be compared against another signature but cannot be turned back into titles. It also holds a fixed random sample of up to 140 titles from your completed list, with the score you gave each: comparing two samples is how the index estimates whether two people rate the same titles alike. All of it comes from public profile data, and a private list contributes no sample.
Reading happens on a queue capped at fifteen AniList requests a minute, which is why a brand-new username can take a moment to appear. If you would rather not be listed, you can ask for your username to be removed.