The honest headline: across 16,738 graded predictions, the team we favoured won 57.7% of the time. If you have seen prediction tools advertise 90%, that sounds low. It is also a number you can check, and the more useful findings are inside it.
Every prediction on FaceitScout is stored before the match starts. When the match ends we record what happened and never touch the prediction again. This is that ledger as of November 2026.
Both are graded the same way: the prediction for the map that was actually played, against who actually won. A match that appears in both is counted once.
FACEIT's matchmaker exists to build even games, and it is good at it. Most lobbies really are close to 50/50, and no model can predict a coin flip. So the fair question is not "can you predict FACEIT matches". It is "how much edge over a coin flip can you actually find, and where".
The real test of a probability is calibration: when we say 65%, does that team win about 65% of the time? Here is every graded prediction, grouped by the chance we gave one of the two teams.
| We said | On average | They actually won | Predictions |
|---|---|---|---|
| under 10% | 4.9% | 12.2% | 156 |
| 10–20% | 15.5% | 10.5% | 172 |
| 20–30% | 25.5% | 26.7% | 300 |
| 30–40% | 36.5% | 34.3% | 1,035 |
| 40–50% | 46.0% | 45.4% | 6,544 |
| 50–60% | 53.9% | 53.4% | 6,674 |
| 60–70% | 63.4% | 64.4% | 1,190 |
| 70–80% | 74.6% | 75.7% | 309 |
| 80–90% | 84.9% | 82.7% | 173 |
| over 90% | 95.3% | 94.1% | 185 |
Read across any row. When we said about 64%, teams won 64%. When we said about 75%, they won 76%. When we said about 46%, they won 45%. From 20% to 90% the prediction and the result are within about two points of each other.
Three things to take from it:
In July we published the first read of the member ledger at about 800 matches. Some of those numbers have come down since, and we would rather show that than quietly keep the old ones.
| Measure | July | November |
|---|---|---|
| Members who played the map we recommended won | 60.6% (132 matches) | 55% (271 matches) |
| Members who played any other map won | 48.1% | 48% (1,917 matches) |
| Calls of 60% or more that hit | 76.7% (129 calls) | 71% (3,550 calls) |
This is what usually happens when a small sample grows: the early figures were flattered by luck, and more matches pulled them towards the truth. A figure built on 3,550 calls is one we expect to still be standing next year. The one built on 129 was not.
You cannot aim better in the next ten minutes. You can choose the map. In the matches members scanned, lobbies that ended up on the map we recommended won 55% of the time against 48% elsewhere, and when that recommendation carried a real edge (55% or better) it was 66% across 119 matches.
Two cautions, because the point of this page is that you can trust it. The recommended-map group is 271 matches, so that 55% carries a margin of roughly six points either way. And we cannot fully separate "the map helped" from "teams that get their map are the teams that win vetoes". The direction has held since March. The exact size is still settling.
The same figures are on our front page and update by themselves as matches finish. Our front page no longer quotes a fixed accuracy number: in November we found it still showing July's figures, so we rewired it to read the live ledger.
See what the model says about your own last match. No account needed.
Scan my last match