Building my own chess analyzer

I did not need another chess product. I needed a way to sit with a game I had just played and see the moments I actually misjudged.

Most tools I tried were loud. Evaluations stacked on evaluations. Lines I would never play. A feeling that the software was more interested in being complete than in helping me think.

So I built a small analyzer of my own.

The actual problem

After a session I already know, roughly, where the game turned. What I cannot reconstruct is why a move felt right at the time. An engine can tell me I was worse. It cannot tell me what I believed.

The analyzer had to answer a narrower question: what did I think was happening, and what was actually happening?

Architecture

The board lives in the browser. The engine runs in a worker. Games I want to keep sit in a thin store. That is the whole system.

type Evaluation = {
  depth: number;
  score: number;
  line: string[];
};
 
export function interpret(eval: Evaluation) {
  if (eval.depth < 12) return "too shallow to argue with";
  return eval.line.slice(0, 4).join(" ");
}

The interesting code is not the engine. Stockfish already exists. The interesting code is the restraint around it.

Implementation

I started with a board I could click without thinking. Then a PGN paste. Then a single engine line, not a forest of them.

The first version showed too much. I removed the extra lines. I made the evaluation a secondary voice — present, but not the headline.

That decision did more for the product than any feature I added later.

Lessons

A personal tool fails when it tries to impress its owner. It works when it disappears into the hour after a game.

I still open this one. That is the only metric I trust.