Wavu Wank

About

This website is still in early alpha and may change dramatically or have significant downtime.

It's inspired by ratingupdate.info, which more or less does the same thing for Guilty Gear: Strive.

Data

The server queries the TEKKEN 8 replay API every ~60 seconds for the latest 999 ranked replays and saves them in a database. This uses the same endpoint that the game uses when you go to the "Online Replays" tab.

Replay data starts from about Mar 27 2024, which is about when this project was started. It's not possible to query the API for old replays, and they seem to be deleted with new patches.

There are some gaps in the data from when the API query script stopped working. There may be more gaps in future. Check the status page to see if things are not currently broken.

For a sense of scale to this data, as of writing after about 12 days there are ~20 million replays in a ~4GB database. (At some point I will have to come up with a strategy for pruning this data.)

Ratings

Ratings are calculated using glicko2, which is one of the most well-regarded and battle-tested rating systems out there. This is in stark contrast to TEKKEN 8's rating system, which is trivially gameable by cherry picking.

Each player gets a completely new, distinct rating for each character they play. Because of how glicko2 works, it will quickly (within ~10 games) reach a reasonable rating, so there's no need for anything like the in-game auto promotion of side characters.

Leaderboards

Ratings with a deviation lower than 75 can appear on the leaderboard. This is what lichess uses. There's no other limitation currently.

I'm aware that there are some players on the leaderboard currently who are obviously cheating. They'll be removed in due time.

Limitations

These ratings are far from perfect. The main limitations are:

Ratings between regions can't be compared. Dominant results in stronger regions are harder to achieve, so top players in stronger regions are expected to have lower ratings than those in weaker regions.

Regions could be more accurate. Regions are currently based on what the TEKKEN game server decides your "area" is. This does not necessarily correspond to who players play against. I am manually moving players who are clearly in the wrong region (i.e. all their opponents are from another region), but I can't realistically do this for everyone.

There's also a glaring flaw: the entire continent of Africa just... doesn't have its own region.

I'm looking into a solution that determines regions by looking at match data, but doing it at the scale of data we're dealing with here is difficult.

People can plug to avoid taking losses. My data is limited to what the TEKKEN game servers record. Until they start recording disconnects as losses, this will continue being an easy way to get fraudulent ratings.

Plans

There are a lot of cool stats, visualizations, etc. one could make from this data. Here's a short list of some things that are planned:

I may or may not make the website look prettier at some point. Current priority is on adding new features.

Who

This site is made and run by 6 weetbix.

As you might expect from the name and URL, I also run wavu.wiki, which you should definitely check out (and contribute to).