Esports Predictions Powered by Analytics

A series read that loses by exactly one map — 1–2 in a best-of-three, 2–3 in a best-of-five — is graded as a push, not a loss. The rule is stated on every esports page because it changes how the published hit rate should be read.

The order matters. Because the market-implied chance is derived from the line before the model writes a word, the breakdown cannot talk itself into a number the market does not support. What the model adds is context — the best-of format, recent series form, roster changes, the stage of the tournament — and an honest account of where the public feed is thin. Per-map win rates and player ratings enter the breakdown only where bo3.gg carries them, and there is no weather and no props; a breakdown that leaned on inputs it does not have would be inventing them, and the pages here do not claim them.

Schedules, live scores, results, rosters and tournaments come from public esports feeds (bo3.gg, with OpenDota and Leaguepedia as reserves); real lines come from a free public bookmaker feed where it covers the match, otherwise the price is a model estimate marked “est.”. The model estimate exists so the board can price a series the book ignores; it stays inside the product, and the public record labels ROI accordingly. Selection is not “take the favourite”: a read is placed when its expected value at the available price clears the floor, after the model’s series edge has been shrunk by a calibration factor, which is why some heavy favourites are passed over.

The grading rules shared with the sports side — one calendar month in UTC, hit rate with pushes excluded, ROI at a flat stake — are written up on the methodology page.

Read these numbers with the same scepticism you would apply to anyone else's. Past results do not predict future results: a strong month is partly variance, and a sample of a few dozen series says very little about a method. Three things specific to esports change how the tables should be read. First, a series lost by exactly one map is a push, so the hit rate runs higher than a plain win-loss count would — the W-L-P column is the honest part of the story. Second, ROI is measured partly at model-estimated prices, because the free public book covers the top divisions and not every series on the board; a month heavy in unpriced series leans harder on est. prices.

Third, coverage is partial by design: the public book prices some divisions and not others, the feed can lag a tournament, and a series without a book line shows no percentage here rather than a made-up one. PUBG and Fortnite have no row because there is nothing to grade. Small samples move fast; wait for volume before drawing conclusions, and treat any month with only a handful of settled series as anecdote.