The engine can't teach you. Here is what a question does that +1.3 cannot.

+1.3. That's what the engine gives you. It's accurate to a tenth of a pawn and it shows up in milliseconds. It says nothing about you. It doesn't know why you played the move or what you were going for, it can't tell which of your habits got you there, and it won't notice when you do the same thing next week.

This post is about the gap between a grade and actual feedback. It's also about why a coach asks you a question before telling you anything. The question is where most of the learning happens.

If a teacher has read one review of feedback research, it's probably John Hattie and Helen Timperley's. Their main claim is simple enough. Feedback is among the strongest influences on learning, but how much it helps depends entirely on what it's about. Feedback on the task, on your process, and on how to regulate yourself next time works. Feedback aimed at you as a person (“good job”) does almost nothing. Good feedback, they argue, answers three questions: where am I going, how am I going, and where to next.¹

A later meta-analysis put the average effect lower than Hattie's early syntheses did, and found it varied hugely, for the same reason: most feedback isn't about the process.² An eval answers “how am I going” in about the narrowest way possible, and the other two stay empty. It scores the outcome of a decision without ever seeing the reasoning behind it.

Nate Kornell, Matthew Hays and Robert Bjork had people guess an answer before they were shown it, and set things up so the guess was almost certain to be wrong. Even so, the failed guess helped people remember the right answer later, compared with just studying it.³ Janet Metcalfe's review of the research on errors lands in the same place from a lot of different angles. Making an error and then getting corrected beats being handed the answer, and instruction built to avoid errors is likely counterproductive.⁴

In chess terms, say you commit to a reason: “I took on e3 because trading off that knight looked like the simplest way out.” Now there's something for the correction to hook onto. When the coach shows you that Black recaptures with check and picks up the knight on f7 a move later, you know exactly which thought it's correcting. Compare that with the engine just showing you Qe6 when you never committed to anything. You nod and move on, and none of it sticks.

The confident mistake

18. Nxf7 Ne3 19.Bxe3

MovesSummary

You spent a real think on Bxe3. What was your evaluation of the resulting trades, and what other candidate moves were you comparing it against?

You

This seemed to be the simplest reply to prevent the fork on the rook and the queen by exchanging pieces. I get rid of that dangerous attacker, and I could not see an immediate mating sequence that Black is threatening.

After 19...Qxe3+, Black has the forcing check that turns your intended exchange into a material loss: 20.Kd1 Nxf7 21.Rc1 Kh8 22.Re1 Qf4 23.e5. You correctly saw that Bxe3 stops ...Nxd5, but the queen recapture with check comes first. Before any capture, list your opponent’s checks, then their captures, and follow the line until the exchanges have ended.

Try this instead:Qe619.Qe6 was the tactical resource that kept the initiative, attacking the knight on e5 while setting up the exchange on f7.

The blunders you were surest of are the easiest to fix

Brady Butterfield and Janet Metcalfe expected high-confidence errors to be the hardest to correct. They found the opposite. On general-knowledge questions, the wrong answers people felt surest about were the ones most likely to get fixed on a later test, once they'd seen the right answer.⁵ This is called hypercorrection. It needs corrective feedback to happen, and it's mostly been shown with trivia, so we don't want to push it further than that. It suggests the move you were sure about, and wrong about, is the one you're most likely to stop playing, but only if someone gets you to say how sure you were and then shows you the fix.

That's why a coach asks how sure you were. The engine never asks.

Thinking about thinking

18. Nxf7 Ne3 19.Bxe3

MovesSummary

You spent a real think on Bxe3. What was your evaluation of the resulting trades, and what other candidate moves were you comparing it against?

You

This seemed to be the simplest reply to prevent the fork on the rook and the queen by exchanging pieces. I get rid of that dangerous attacker, and I could not see an immediate mating sequence that Black is threatening.

After 19...Qxe3+, Black has the forcing check that turns your intended exchange into a material loss: 20.Kd1 Nxf7 21.Rc1 Kh8 22.Re1 Qf4 23.e5. You correctly saw that Bxe3 stops ...Nxd5, but the queen recapture with check comes first. Before any capture, list your opponent’s checks, then their captures, and follow the line until the exchanges have ended.

Try this instead:Qe619.Qe6 was the tactical resource that kept the initiative, attacking the knight on e5 while setting up the exchange on f7.

Did you notice Qxe3+ was waiting for your opponent after this?

I did not see this replyI saw it but thought I was still okayI saw it but calculated the line wrongNot sure / don’t remember

The skill engine review never trains

John Flavell named it in 1979: metacognition, knowing about and keeping track of your own thinking.⁶ A player who's good at this knows when they're calculating and when they're guessing. They notice when their plan has quietly shifted. They can tell “this move is good” apart from “I want this move to be good.” None of that is chess knowledge, and all of it shows up in your results.

You can't practice keeping track of your thinking if nothing ever asks you to report it. A question makes you do that. You have to look back at what you actually did and put it in words, and that's usually when you find out it wasn't what you thought.

Strong players get something out of slowing down too. One study scored moves with engine evaluations and found that both strong and weaker players picked objectively better moves after extra deliberation than on their first instinct. That held on easy and hard problems alike.⁷

We're not saying passive review is useless. In the classic testing-effect study, students who reread a passage actually beat the self-testers on a quiz five minutes later. The flip only showed up a week out.⁸ The engine is the most accurate judge of a chess position there has ever been, and Chessalyz runs one under everything. What we changed is the order. You go first. The engine checks your answer after.