Why a chess coach shouldn’t be allowed to invent the position

You paste in a game, ask for an explanation, and receive something that sounds perfectly sensible:

“You missed a tactical opportunity to attack the pinned knight and open the h-file.”

The only problem is that there was no pinned knight. No h-file attack. Sometimes, no tactical opportunity at all.

The explanation has the familiar rhythm of chess language, so the problem may not be obvious at first. But once you notice it, it becomes difficult to trust the rest of what it says. A chess coach does not need to be brilliant all the time. It can say that it is unsure. It can focus on a simple mistake. It can point out that a move lost a pawn because you missed a line. But it should not be allowed to invent a story about the board and present it with confidence as though it had actually happened.

So I built a chess review tool around one fairly strict idea:
The explanation should follow the analysis, not replace it.

The gap between analysis and explanation

A chess engine is not a coach. Stockfish can tell you that one move is much better than another, show a principal variation, and evaluate the position. But most people do not want to review a whole game through raw engine lines alone.

A language model, on the other hand, can turn ideas into clear explanations. But a clear explanation is not necessarily a reliable one.
Those are two different strengths, and neither one alone was quite the experience I wanted.

I wanted a review that could make the important moments of a game easier to understand without losing track of the position itself.
That became the standard I set for the project. It took much more work than the final experience suggests, but the difference is easiest to see in a real position.

Not every move needs to sound deep

Many chess explanations become less useful when they try too hard to sound profound.

Sometimes the lesson is simply that a pawn was hanging.
Sometimes the best move is a quiet recapture.
Sometimes a move is neither a blunder nor brilliant. It is not part of a grand plan. It is just fine.

I wanted the review to leave room for that. Not every position needs a dramatic story. If a move fails because of one tactical detail, then that detail is enough. If the engine prefers a move because it keeps a small advantage, that should not be presented as a forced win.
The point is not to make an explanation sound fancy. It is to make the important ideas easier to understand.

What I was able to achieve

Here is what that standard looks like in a real position.

Position after 34. Rf8+. Black to move.

Here is the full explanation from the review:

You played rook to f8 with check, and the opponent moved their king to h7. You then played bishop to d3 with check, the opponent pushed their pawn to g6, and you delivered checkmate with queen to f7. While this sequence ended in victory, it relied on the opponent choosing king to h7 instead of the strongest defensive resource.

The critical problem is what happens if the opponent simply takes your rook with rook takes f8. Your pawn on e7 could promote to a queen, but the opponent’s rook would immediately capture it. The result is a clean trade of rooks where your advanced pawn vanishes without creating any material advantage. This refutation neutralizes your entire combination, turning a winning position into one where you have surrendered your key asset for nothing. The opponent’s failure to play rook takes f8 was the only reason your attack succeeded.

A much stronger plan was bishop c4 to b5. With this move, you would attack the opponent’s rook on e8 directly. The opponent would likely respond by moving that rook to c8 to escape the threat. You could then reposition your queen to d6, and the opponent might counter by moving their queen to d2, attacking your rook on e1. This sequence maintains material balance but keeps your pieces actively coordinated, preserving your winning chances without risking your passed pawn.

The key calculation habit here is to always check your opponent’s most forcing capture before committing to a combination. Before playing rook to f8, you needed to ask: what is the simplest way for the opponent to take this piece? By visualizing rook takes f8 and the subsequent pawn promotion and recapture, you would have seen that the combination loses its punch. This discipline of testing the most direct reply first would have revealed the flaw and pointed you toward the safer bishop move.

This is the kind of explanation I want from a chess coach. It starts with what happened in the game, tests the obvious defensive resource, explains why the attack fails, and gives a concrete alternative.

Most importantly, the important claims are possible to check on the board.

What I am still working on

This is not a solved problem.

Quiet positions can be difficult to explain. An engine line can be technically correct without being a good teaching tool. And seeing numbers does not automatically mean that an explanation is helpful to a human player or that they would be able to fully understand it.
That is the part I mainly want to improve.

When you review your own games, what helps most? Seeing the critical position again? A short explanation of the tactical point? One good line that is easy to understand? A summary of recurring habits after the game?

I think chess tools work best when they make you want to return to the board and look again, not when they simply give you a verdict to accept.

This idea became Zen.

Zen is not just a page that analyses a game and then forgets it. It is a browser-based chess workspace with a dashboard where you can keep your games, return to them later, and build a personal review library stored locally in your browser. Alongside reviewing the critical moments, you can explore positions and variations in Explore mode, use simulations to see how an idea might develop, and try the coaching approach described above.

The core review flow does not require an account. There are other features to explore too, but I would rather let people discover those than turn this article into a feature list.

You can try it here: https://zen.reinit.net