The wall was 1800.

My son Dovi plays chess.

For a long stretch, he was stuck at 1800 Rapid on Chess.com. Not falling, not rising. Drawing games he should have won. Losing games he should have drawn. The kind of plateau every improving player eventually hits.

Here's what's in the toolkit already.

A coach. Helpful. A good coach finds ideas a kid can't find alone, and Dovi came away from every session sharper. But a coach sees maybe eight of your games a month, and only a few of those will contain the specific situation where your weakness shows up. That's not enough data to spot a pattern, especially one that only bites you some of the time. The habits costing you rating live in the hundreds of games in between, and no coach has the hours to watch those.

Books. Silman's endgame book helped. Dovi knew his endgames needed sharpening, and it delivered. But his real weakness wasn't the endgame. It was the middlegame, where unsound ideas kept panning out often enough to feel like style. No book helps with a weakness you don't know you have. That's the chicken-and-egg nobody talks about: improving players don't know what they don't know.

Here's the conversation that got me writing code.

Dovi and I were talking about one of his losses. He'd launched an attack on the opposing king too early. When the attack fell apart, he was left with a position that was easy to exploit, and he got ground down. I said something like, "you do this a lot." He pushed back. He'd won games doing it. And he was right. He had. About 60% of the time, especially against weaker opponents, especially in blitz. It looked to him like style, not a weakness.

That was the thing.

The pattern was invisible to him because it mostly worked. Partial reinforcement. A move that wins more often than it fails doesn't feel like a weakness. It feels like who you are.

A human can't reliably spot a pattern that only hurts them sometimes. A system looking across twenty of their games at a time can.

So I built one.

Promote pulls your games from Chess.com or Lichess, twenty at a time, and keeps syncing as you play. It runs Stockfish on each one. It finds the patterns that show up across games, not just within one, and it trains them back to you using positions from your own games, not generic puzzle sets.

As of the time of this writing, Dovi is at 1870 Rapid. I can't run a proper A/B on my own kid, so I won't claim Promote is the only reason. But the wall moved, and he stopped losing to the same pattern over and over. When he launches an attack now, it's because the position actually supports it.

None of this replaces his coach or his books. Dovi still has both, and he still needs both. They're the tools that teach you how to play better chess. Promote does something neither can: look across a batch of your games at once and find the patterns that don't show up in any one lesson or any one chapter. A new tool in the kit, not a replacement for the rest.

I'm releasing Promote because there are a lot of Dovis out there. 1400, 1800, 2000. It doesn't matter where the wall is. Every improving player has a blind spot that sometimes works, and a tool that can look across enough of their games at once can find it.

— Daniel

P.S. If you want to be one of the first hundred players on the site — lifetime free, pay only for AI tokens — take a look at Founding 100.