About SaltyTrack

The story behind the data

Who We Are

SaltyTrack was created by A. — a data enthusiast, hobbyist coder, and longtime SaltyBet viewer who got tired of guessing and started counting.

What started as a side project to track a few matches turned into a full-blown analytics platform. Since late 2021, SaltyTrack has recorded over half a million matches, cataloged 10,000+ unique fighters, and built a machine learning prediction model that outperforms the crowd.

Why SaltyTrack Exists

SaltyBet is a 24/7 AI fighting game betting platform with thousands of characters, and the only way to know who’s good is to watch thousands of matches — or have someone track them for you.

SaltyTrack does the tracking. We collect every match result in real time, compute fighter statistics, and run them through a trained ML model to generate predictions with confidence levels. The Chrome extension overlays this data directly on saltybet.com so you see it during the betting window.

Our Data

Everything on SaltyTrack is built on real match data collected from the SaltyBoy API. We don’t scrape, guess, or make things up. Every statistic, win rate, and prediction on this site traces back to recorded match outcomes.

Our prediction model consistently scores in the low-to-mid 70s for overall accuracy, and at its highest confidence level (Lock) it sits close to 99%. We’d rather you didn’t take our word for any of that: the Accuracy Dashboard publishes the current numbers, breaks them down by confidence level, and scores the model head-to-head against the betting crowd over the exact same set of matches. Those numbers move as new matches are scored, and they include the misses.

For the full story of how this project came together, read Why I Built SaltyTrack.

How We Build the Numbers

Every figure on this site comes out of the same four steps, and it’s worth knowing what those steps actually do:

  1. Collect. A collector watches SaltyBet around the clock and records each match as it settles — the two fighters, their tiers, the Salty Bucks wagered on each side, and who won. Nothing is typed in by hand.
  2. Aggregate. Win rates, tier records, win streaks, and head-to-head history are computed from those stored results. They’re counted, not estimated.
  3. Predict. A machine learning model trained on the archive turns a matchup into a win probability, which we bucket into confidence levels — Lock, High, Medium, Low. How SaltyTrack Predicts Fights walks through what it considers.
  4. Score. Every prediction is stored with its match and then graded against the actual winner. The model is handed nothing but the two fighters’ names and the record built up before that match — never the result it’s about to be scored on. That’s what makes the Accuracy Dashboard an audit rather than a brag.

Coverage: late 2021 through today, and still counting. Cadence: matches are recorded as they finish, and site statistics refresh within minutes of a result.

And what we can’t tell you: the model only knows what has been recorded. A brand-new fighter with three matches to their name is close to a coin flip no matter how confident the interface looks, and tier changes, roster edits, and joke characters all add noise. We publish where the model struggles next to where it shines — those limits are the whole reason the accuracy page exists.

Contact

Have a question, found a bug, or want to say hi? You can reach us through the GitHub issues page.

SaltyTrack is not affiliated with, endorsed by, or connected to SaltyBet. All betting on SaltyBet uses fictional currency with no real-world value.