The Chess Improvement Loop Most Amateurs Skip

Here's the uncomfortable thing about how most amateurs "study chess": they do one quarter of the work and wonder why they've been stuck at the same rating for two years.

You drill an opening. Maybe you grind some puzzles. Then you play, lose, shrug, and queue up the next game. The losing game — the one with all the information in it — gets closed without a second look. You never find out why you lost, so you can't train the thing that's actually costing you points. You just hope the opening prep carries you.

It doesn't. Opening prep gets you to move 10 in a position you can play. Everything after move 10 is where your rating actually lives, and you're not looking at it.

This is a post about the full loop: play a real game → review it → see where you stand against players at your level → drill your actual weaknesses → play again. It's the loop every coach teaches, and you can run all of it on MyChessBot without paying anyone $60 an hour.

Step 0: opening prep is the warm-up, not the workout

Don't throw out the opening drilling — it's real, and it's still where you should start. If you keep walking into the same disaster on move 6, fix that first. Pick the opening that wrecks you, pick a bot calibrated to your level, and play the matchup until it stops wrecking you. That's the original MyChessBot pitch and it still works.

Take the Italian Game. Four moves and you've got a sound, classical position you actually understand:

But here's the trap. You can drill the Italian until move 8 is automatic and still lose every game on move 25, because you hang a rook in a won endgame, or you panic the instant you're attacked, or you can't convert a winning position to save your life. The opening was never the problem. The problem is everything you've never measured.

That's where the rest of the loop comes in.

Step 1: review the game you just lost

The single most-cited habit in serious chess coaching — Botvinnik, Heisman, the ChessDojo crowd, all of them — is review your own games. Not grandmaster games. Not the engine's top line in the abstract. Your games, the ones you just lost, with your name on the blunders.

Most amateurs skip this because doing it by hand is tedious and they don't trust their own evaluation anyway. So we built it in. Every game you play gets a full engine review: an evaluation bar down the side, every move tagged the way you're used to from the big sites — best, good, inaccuracy, mistake, blunder — and the engine's preferred move drawn right on the board next to what you actually played.

The point isn't the pretty graph. The point is the gap between the move you played and the move that was there. Our "Review & Improve" mode leans all the way into this: it replays the turning points of your game and asks you to find the move the engine preferred — active recall on your own mistakes, which is the highest-value training rep in all of chess. You don't learn from being told the answer. You learn from blanking, guessing wrong, and then seeing it.

Do this for ten games and a pattern shows up that you could never see one game at a time. Maybe every loss has a blunder between moves 20 and 30, right when the position gets tense. Maybe you're fine until the queens come off and then you have no idea what to do. One game, that's noise. Ten games, that's a diagnosis.

Step 2: find out where you actually stand

Here's the part nobody else gives you straight. Knowing your moves were "62% accurate" tells you almost nothing — accurate compared to whom? A 62% game is excellent for a 900 and a disaster for a 1700.

So we built a baseline from a large body of real, rated games and measure you against the players in your exact rating band. The result is the Skill Profile: more than a dozen parts of your game, each ranked against players at your level — overall move quality first, then, where it actually counts, broken out:

• By phase — opening, middlegame, endgame. Is your endgame dragging the rest down?
• By situation — converting a won position, defending a lost one, and accuracy in sharp, tactical positions. Different skills, and most players are lopsided across them.
• Composure — the one almost nobody else surfaces: whether your play holds up right after a mistake, or you tilt and the next move craters too. Now you know, and "treat the next move as a fresh position" becomes something you can actually practice.

Instead of "you're a 1400," you get "you convert won positions in the 70th percentile for your level, but you defend lost ones in the 20th, and your move quality craters right after you blunder." That last sentence is a training plan. The vague one never was.

You can see a full sample Skill Profile here — real layout, fictional 1500 player — without signing up for anything.

Step 3: train the actual weakness, not random puzzles

Now the loop closes. You've got a ranked list of your real, measured weaknesses. The built-in coach turns that into a curriculum — and "curriculum" is the right word, because it is deliberately not just a stack of puzzles.

A few things make this different from the puzzle-rush trainer you've used before:

It drills your weaknesses, on a spacing schedule. The mistakes you actually make — pulled from your reviewed games — become spaced-repetition drills. Critically, the scheduling unit is the pattern, not the puzzle: when a fork weakness comes due, you get a fresh fork position you've never seen, not the same one again. Re-showing the identical puzzle trains recognition; a new position trains the thing you need in a real game — retrieval.

It aims for ~85% success, on purpose. There's good evidence (Wilson et al., 2019) that learning is fastest when you're getting things right about 85% of the time — hard enough to stretch, easy enough to stay in it. The difficulty self-tunes toward that band. It will not let you farm a vanity rating on puzzles three classes too easy for you.

It's more than puzzles. Depending on your rating and your gaps, the plan mixes in guess-the-move on your own games, "convert this won position" play-outs against a bot, rating-appropriate endgames drilled until they're automatic, and that composure habit cue. A 900 and an 1800 get structurally different plans — the way an actual coach would teach them, not the same puzzle set at a different difficulty.

And the plan isn't a dead-end checklist. Finish a due item and it advances on the forgetting curve; the next-ranked weakness surfaces. It's a living queue that keeps pointing at whatever is currently costing you the most.

The honest part: does any of this actually work?

Most training products quietly dodge this question, because the honest answer in the research is uncomfortable: puzzle-to-real-game transfer is not well demonstrated. Tactics trainers love to imply "solve 10,000 puzzles, gain 300 points," and there's no good study behind it.

So rather than make a number up, we measure it on you. For each pattern you train, we track its mistake rate in your real games over time, split at the moment you started training it — and we compare it against your untrained patterns as a within-you control. If your fork mistakes drop while your untrained patterns stay flat, that's a real signal that the training is doing something. If it's just noise, we say so: "no clear change yet — keep training and playing."

That's the moat, honestly. Anyone can sell you puzzles. We'll show you whether they're working, even when the answer is "not yet." It's a falsifiable claim about your own improvement, which is a thing almost nobody in this space is willing to put on screen.

Running the whole loop (a concrete week)

Enough theory. Here's the loop as an actual routine:

1. Play three slow games against a bot near your level. Positional Pete (1400) is a great sparring partner for this — he won't hand you anything, he squeezes endgames, and he punishes loose structure, so your real weaknesses surface fast instead of getting bailed out by a blunder.
2. Review all three. Don't just skim the eval bar — run Review & Improve on the turning points and try to find the move before you reveal it.
3. Open your Skill Profile once you've got a handful of reviewed games. Read the one dimension where you're furthest below your level. That's your target.
4. Do one focused training session built from that weakness. Fifteen minutes, mixed, at the right difficulty.
5. Play again. Watch whether the thing you trained shows up less.

That's it. The openings get you a playable game; the review tells you what went wrong; the Skill Profile tells you whether it's a you problem or just one bad game; the coach drills it; and the transfer measurement tells you, over weeks, whether the needle is moving.

Most amateurs run step 1 and call it studying. The players who actually climb run the whole loop. The tooling to do it used to cost a coach's hourly rate — now it's sitting in a browser tab, and most of it is free.

Start where it hurts: drill the Italian Game against Positional Pete, then review the game instead of closing it. That one change — looking at the loss instead of queueing the next one — is the difference between two years of plateau and actually getting better.