Dota 2 betting signals you can actually audit, not just promises

No. dota2llm is an analytical tool: it sends probability-based signals and explanations on Telegram. You decide whether and how to act on them. It never places real bets for you.

NO_BET is a deliberate, first-class outcome, not a failure of the model. It shows up most often for a team with only a handful of games in its history — too little track record yet for the model to be confident — rather than the model guessing anyway.

A set of statistical models, one for each stage of a map, trained and validated on thousands of tracked games. The exact methodology isn't published, but every prediction it produces is shown here with its real, ongoing track record.

One of the model's live re-evaluation points during a Dota 2 map — the moment it has enough real game data (net worth, kills, tower state) to re-check its read on who's ahead. The 19- and 28-minute checkpoints have the deepest sample sizes, which is why their live track record is shown on this page.

$25/month, or $7 for 1 week if you just want to try it short-term. The first 7 days are free for new users, no long-term commitment. The live track record above is free to check at any time — you only pay if you decide the model's edge is worth following.

An experimental, still-in-testing secondary signal shown on every live alert: the real win rate of each team's five drafted heroes for its own players. A one-time check run on 2026-10-25, across 3,021 settled 19-minute picks (3,004 with usable data at the time), found a 72.5% real win rate when it agrees with the model's pick, versus 59.2% when it disagrees. It's still being trained and validated on the local statistical model, and isn't used to filter or skip any bet today. The guide also tracks a genuinely live, forward-only version of this same stat, growing day by day from 2026-10-25 onward.

Just Dota 2, and only a single map's winner. dota2llm doesn't cover CS2, League of Legends, Valorant or any other esports betting market — each of the three local models is trained and validated on Dota 2 data specifically, and stretching that accuracy to a different game without retraining it would be dishonest.

Yes, that's the main use case: both live checkpoints (the 19-minute and 28-minute mark) score the map once it's already in progress, exactly when a bookmaker has a live price on that side. Every signal arrives with the current odds and the model's own probability — comparing the two and deciding whether to bet is on you.