Engine ideas

Every position here comes from a real blitz game, at the exact spot where players at a rating reliably choose one plan and Stockfish argues for another. Find the engine’s move, and the coach explains the gap once you do.

51 drills from three phases of the game

Programmers operating ENIAC, c. 1946 · Unidentified U.S. Army photographer

43% of 1100s miss 2.1 pawns

NextDrill 1 · 1100 · different plan

White to move. Find the move Stockfish wants, not the one the crowd plays.

Start drill

Opening

16 drills

Middlegame

31 drills

Endgame

4 drills

How these drills are made

Maia, a neural network trained on millions of human games, says what players at each rating actually play, and how often. Stockfish says what the move really costs and what it would rather do. When the two disagree about the whole plan, not just one move, the position lands here.

You get the board and one fact: the share of players at a rating who fall for the same pull you are about to feel. Play the crowd’s move and you hear why it falls short, with the punishment line. Play anything else and Stockfish prices it on the spot. Hints point at what the position is about: the weak squares, the open files, the backward pawns.

The lesson is the disagreement. When a third of players attack on one wing and the engine plays on the other, the gap between those two stories is the thing worth learning. Every sentence here was checked against the board before it was published.

Batch 1: 51 positions from 1,107 Lichess arena games, filtered from 48,169 candidates. Human probabilities by Maia-2 at 1100, 1400 and 1700; every cost measured by Stockfish at depth 22 and re-verified independently before publishing.