Home / jjLABS / jjB!B / Accuracy check
We checked our own numbers.
When a new man turns up in a lineup, jjB!B sends an alert before tip-off, with a points estimate: what he should score if he really plays. In its first week, from 31 October to 5 November 2026, I went back and checked every one against the final box score. The good calls, the bad ones, and what I changed.
A note on timing: this check uses data from the first days of November, a few days before this page went up. The formula may already be better than what you see here. The real, dedicated work on the projections, and a proper update on it, comes later.
CHECKED
46 new men
36 games, 11 leagues. 40 of them got on the court.
HOW EARLY
20–25 min
before tip-off, in the alerts I checked by hand.
The near-perfect calls
One game, four new men, four alerts sent 20 to 24 minutes before tip:
| Player | Our estimate | Real points |
|---|---|---|
| White | 9.9 | 10 |
| Bernard Louis | 10.8 | 10 |
| Sylla | 17.0 | 15 |
| Davis | 11.5 | 17 |
Three of the four within two points. Davis beat us by five and a half. That happens too.
In the European cups and the NBA:
| Player | Our estimate | Real points |
|---|---|---|
| Kraag | 6.7 | 7 |
| Davidson | 8.4 | 8 |
| Cleary | 9.3 | 10 |
| Strus | 8.8 | 11 |
The misses, and why
Across all 40 men who played, about half landed within 5 points of our number, and 30% within 3. The rest missed, some by a lot.
The pattern was clear. Almost every big miss was a man his coach barely used. All 11 who played under 10 minutes scored less than we said. The men who played 20 to 30 minutes were off by only 3.7 points on average.
That is the honest limit of a number like this. It tells you what a man does when he really plays. Nobody knows his minutes before the game: not me, not the software, not anyone.
What I changed
The estimates ran too high, so I rebuilt the formula. Then I re-ran it on the same men in the same games, 151 alerts, to see what it would have said:
| Old formula | New formula | |
|---|---|---|
| Average miss | 6.25 pts | 4.74 pts |
| Average overshoot | +3.82 pts | +1.67 pts |
It also learned to stay quiet. The new formula would not have sent 54 of those alerts. Those men really scored 1.9 points on average. The 97 it kept really scored 9.1.
One example: Bae Byungjun. The old formula sent him as new, at 10.0 points. The new one sees he is back after missing 9 games, estimates 4.1, and stays silent. He scored 0.
This is a beta. People who use it find bugs, and I fix them fast. Checking my own numbers in public is part of the same job.
Not in the box
No picks. No bets. No odds. jjB!B tells you who is in, who is out, and what that man usually brings, before the game starts. What it means for a game is your call.
If you don’t bet basketball, don’t even start.
The beta
jjB!B is in a limited, closed beta on Telegram. Everything about it, including how to take part, is on my Telegram channel.
If betting is already a problem, free and confidential help: BeGambleAware.