Your edge shows up everywhere. It is thinnest in the endgame.
Almost everything you play is blitz, and 65% of it is 3+0. Over the last 90 days your rating went from 3384 to 3394, and your accuracy led your opponents’ 92.7 to 90.2. We ranked six leaks by the rating points each costs you beyond what it costs your opponents in the same games. At all six you lose the same or less than they do. So #1 is not a weakness. It is where your edge is smallest: the endgame, where you and your opponents both give up about 7 points.
The short answer
Analysed: MagnusCarlsen’s chess.com blitz games: 6,902 games · Stockfish 19 on the latest 2,000 (Mar 23 2025 – Sep 24 2026). Report date: .
Verdict: Holding steady at the top. Last 90d vs Prior 9 mo: accuracy 92.7 vs 92.7 (p = 0.97); big mistakes per 40 moves 0.50 vs 0.49 (p = 0.74); score 76.6% vs 76.7% (p = 0.95). Wins from lost positions: 2% vs 4%, so none of it is luck. The last 30 days are sharper: accuracy 93.6 (p = 0.017) and 0.27 big mistakes per 40 moves (p < 0.001), but that is only 73 games.
Biggest leaks:
- Endgames: the last point: 403 endgame mistakes · 49% in rook+minor endings · ≈7 rating points, same as your opponents
- Openings that start you in a hole: White: +0.4 by move 10 · Black: −0.3 · ≈16 rating points, 3 fewer than your opponents
- One-move oversights: ≈119 mistakes · 11% of all · ≈12 rating points, 5 fewer than your opponents
Free Goat Card for MagnusCarlsen · How to review your own games for free · Time pressure in chess, measured
Your game at a glance
- Openings3,110▼ 20
- Tactics3,470▲ 10
- Strategy3,250■ 0
- Endgames3,260▼ 10
- Brilliant moves306in 2,000 games
Each number is the rating of a player who plays that part of the game the way you do: your opponents’ average rating (3,116), adjusted by how much win-% you lose per move there compared with them, in the same games. Arrows: the change against your opponents compared with Prior 9 mo; grey means it is within the noise. A brilliant move is the engine’s top choice that gives up at least two points of material.
Improving, or lucky?
| Metric | Last 30d | Last 90d | Prior 9 mo | Read |
|---|---|---|---|---|
| Results | ||||
| Games | 73 | 258 | 1,061 | |
| Rating, start → end | 3312 → 3394 | 3384 → 3394 | 3334 → 3381 | flat |
| Score | 84.2% | 76.6% | 76.7% | flat (p = 0.95) |
| Skill: what the engine sees | ||||
| Your accuracy | 93.6 | 92.7 | 92.7 | flat (p = 0.97) |
| Opponents’ accuracy | 90.2 | 90.2 | 90.3 | flat |
| Big mistakes per 40 moves | 0.27 | 0.50 | 0.49 | flat (p = 0.74) |
| Opponent blunders you punish | 88% | 73% | 78% | worse |
| More accurate than opponent | 82% | 74% | 78% | worse |
| Eval at move 10 as White | +0.6 | +0.4 | +0.4 | flat |
| Eval at move 10 as Black | −0.1 | −0.5 | −0.3 | flat |
| Conversion & luck | ||||
| Score after reaching +5 | 95.5% | 96.3% | 94.7% | flat |
| Score after being −5 or worse | 0.0% | 11.4% | 19.9% | saving fewer |
| Losses that were +5 at some point | 22% | 10% | 16% | better |
| Wins that came from a lost position | 0% | 2% | 4% | flat |
| Wins on the clock while losing on the board | 0% | 1% | 1% | flat |
| Losses on time | 33% | 15% | 22% | better |
Rating, rolling 50-game average
A slow, steady climb. Rolling 50-game average: 3248 → 3374.Accuracy, rolling 50-game average
The blue line (you) stays above the orange one (your opponents) the whole way. Now: you 93.8, opponents 90.2.Your clock: wins vs losses
Time left on your clock after your move, in the games you won and the games you lost.
- Move 20
11 seconds apart: 2:03 in your wins, 1:52 in your losses. Medians over your 3+0 games (928 wins, 249 losses).
- Move 30
The gap grows to 14 seconds: 1:18 in wins, 1:04 in losses (733 wins, 213 losses). In 3+0 that is the time for a few careful endgame moves.
Nothing in the trend needs fixing. At this level the points left are small and specific, and the leaks below show where. Judge the last 30 days on your next 50 games before you trust them.
What you do well
Not everything is a leak. These parts of your game already work, measured the same way as the leaks below: against your opponents, in the same games.
- Tactics: 3,470354 above your opponents’ average of 3,116: you see and play tactics better than players at your level.
- Tactical defenceShort combinations cost you less than they cost your opponents: ≈28 points, against ≈69.
- Closing out winsYou close out won games better than your opponents do: ≈14 points lost from winning positions, against ≈33.
- Punishing mistakesYou cash in on their mistakes more often than they cash in on yours: ≈22 points lost there, against ≈28.
- Few pieces hungYou hang pieces less often than your opponents: one-move oversights cost you ≈12 rating points, them ≈17.
Your best moves
6 leaks, ranked by rating points lost beyond your opponents
Each leak is ranked by the rating points it costs you beyond what it costs your opponents in the same games. You are level with them or better at every one, so read the list as “where your edge is thinnest”, not “what is broken”. Every position below is from one of your own games in the last 365 days. Decide on your move first. Then open the answer and step through what you played and what was best with the ← → buttons (or your arrow keys).
1 · Endgames: the last point
403 endgame mistakes · 49% in rook+minor endings · ≈7 rating points, same as your opponentsIn order to improve your game you must study the endgame before anything else; for, whereas the endings can be studied and mastered by themselves, the middlegame and the opening must be studied in relation to the endgame.
59% of your games reach an endgame. From a winning one you score 95%. Endgames cost you ≈7 rating points, exactly what they cost your opponents in the same games. Everywhere else you gain on them; here you only break even. 49% of your 403 endgame mistakes come in rook+minor endings. You lost 7 winning endgames on time. Three of the four positions below were played with 21 seconds or less on your clock.
- Why it happens
- Your endgames are sound; they are just played on almost no time. At move 40 your clock reads 0:51, and your median endgame move takes 1.1 seconds. Two of the four positions above were a mate in one and a free knight, both missed with 21 seconds or less.
- The habit
- When the queens come off, spend your time once: name the winning plan and his only counterplay. Then play the technique fast.
- The drill
- Once a week, take one rook+minor ending from your own games where you were winning and play it out against the engine from that position, with the clock you had. For the theory behind them: Dvoretsky’s Endgame Manual and Müller and Lamprecht’s Fundamental Chess Endings.
- What success looks like
- Convert at least 96% of your winning endgames (now 95%).
Practice this: Converting a won game: trade pieces, not pawns →
2 · Openings that start you in a hole
White: +0.4 by move 10 · Black: −0.3 · ≈16 rating points, 3 fewer than your opponentsYou should keep in mind no names, nor numbers, nor isolated incidents, not even results, but only methods.
By move 10 you are +0.4 as White and −0.3 as Black. In 4% of your Black games you are already 1.5 pawns down, against 1% as White. That costs ≈16 rating points, 3 fewer than your opponents lose to their openings. The repeats are easy to price: 1…f5 after 1.e4 Nc6 2.Nf3 (7 games, about 10 win-% each), 1…b5 against 1.d4 (6 games, 9%), 1.f3 and 1.Nh3 (3 games each, 8%).
- Why it happens
- Your openings don’t lose games: you score 76% as Black and 79% as White. The gap is at move 10, −0.3 as Black against +0.4 as White, and a few offbeat first moves you play again and again. Each one costs you about 8–10 win-% in the first two moves.
- The habit
- For every offbeat first move you keep playing, know the critical reply and your answer to it. All four positions above turned on a single point by move 12: a recapture, a count on e5, a defender, a cleared rank.
- The drill
- Engine and database on your worst named lines: the Rossolimo (14 games, 68%) and the Najdorf (−0.29 at move 10) as Black, the Sämisch Nimzo-Indian (15 games, 60%) as White. Follow your own games to the move where they leave theory and check the critical branches at depth.
- What success looks like
- Average eval at move 10 as Black at or above −0.3 (now −0.3).
- Lichess Analysis board + Opening ExplorerpracticeMaster database and cloud engine on one board: enough to check your own lines to the move where they leave theory.
Practice this: Opening rules that win games →
3 · One-move oversights
≈119 mistakes · 11% of all · ≈12 rating points, 5 fewer than your opponentsMistrust is the most necessary characteristic of the Chess player.
One-move oversights are rare in your games: ≈119 in 2,000, 11% of your big mistakes, and they cost you 5 rating points fewer than your opponents lose to theirs. 72 times a piece went to a square where it could be taken; 39 times an attacked piece stayed where it was. The clock explains most of them: 0.6 big mistakes per 100 moves with more than 90 seconds left, 3.7 with under 10. All four positions below came with 37 seconds or less.
- Why it happens
- You see these with time. Your big-mistake rate is six times higher under 10 seconds (3.7 per 100 moves) than above 90 seconds (0.6). The check disappears when the clock gets low.
- The habit
- Under 20 seconds, one question before you let go: “What can reach the square I just moved to, and what did I leave?”
- The drill
- Turn your ≈119 oversights into a flashcard set: each position with 5 seconds to name what the moved piece left loose or walked into. Twice a week, until you get them all.
- What success looks like
- Big mistakes at or below 0.4 per 40 moves (now 0.50), measured on your next 50 games.
Practice this: The blunder check →
4 · Not taking what’s offered
307 missed punishments · 25 free pieces · 4 mates · ≈22 rating points, 6 fewer than your opponentsI have a weakness for any piece in excess of my opponent's numbers - from pawn to queen.
Your opponents hand you material, and 73% of the time you take it (88% in the last 30 days). The misses cost ≈22 rating points, 6 fewer than your opponents lose on theirs. 25 of the 307 were free pieces, 4 were mates. 172 of the winning moves were quiet, neither a check nor a capture, and 73 of those attacked a piece with nowhere to go.
- Why it happens
- In fast games the winning reply is often quiet: 172 of your 307 misses were neither a check nor a capture. The forcing moves get looked at first, and the quiet win is not among them.
- The habit
- When his move looks like an error, look for the quiet move first: the piece with no retreat, the move that takes a square. 73 of your misses attacked a trapped piece.
- The drill
- Aagaard’s Grandmaster Preparation: Calculation, the candidate-moves chapters: two positions a day, written out in full before you check. Then the 307 missed punishments from your games, 30 seconds each.
- What success looks like
- Punish at least 80% of your opponents’ blunders (now 73%).
Practice this: Is it really safe? Counting attackers and defenders →
5 · Giving back won games
Reached +5 in 1,364 games · lost 52 · drew 30 · ≈14 rating points, 19 fewer than your opponentsFirst restrain, next blockade, lastly destroy.
You reached +5 in 1,364 games and lost 52 of them. 42% of your big mistakes come from winning positions, at a median 0:49 on your clock. Your opponents give back more: this leak costs you 19 rating points fewer than it costs them. 32 of the 52 losses from +5 ended on time. Three of the four positions below were forced mates.
- Why it happens
- When you are winning, your opponents’ only chances are a check and the clock. 32 of the 52 games you lost from +5 ended on time, and the median clock when you blunder a winning position is 0:49.
- The habit
- Winning with little time: if you see a forcing finish, play it; otherwise take away their checks. Trade queens when it costs nothing.
- The drill
- Spar the engine at 3+0 from the moment you reached +5 in your own games, the positions above and the 52 losses, with the clock you had. The task is to win without allowing a check through.
- What success looks like
- Score at least 97% after reaching +5 (now 96%).
Practice this: Converting a won game: trade pieces, not pawns →
6 · Two-move tactics against you
≈377 mistakes punished by a short combination · ≈28 rating points, 41 fewer than your opponentsStrategy requires thought, tactics require observation.
Short combinations against you: checks (98), capture sequences (95), attacks on pieces with no retreat (71), forks (49). This is your largest cost in raw points, ≈28, and also your biggest edge: your opponents lose 41 more rating points to the same thing in the same games.
- Why it happens
- These combinations are short, and most start with a forcing move: 98 of them with a check, 95 with a capture sequence. In the positions above, three of the four started with a capture of your own: …exf5, Bxe7, Qxd4.
- The habit
- Before a capture in a sharp position, ask “What is their most forcing reply?” Check it first.
- The drill
- Composed studies and the opponent’s-resources chapter of Aagaard’s Calculation: in each position, find the defender’s most forcing reply before you choose your own move. 15 minutes, twice a week.
- What success looks like
- At the next audit, no more than 23% of your big mistakes come from a short combination (now 35%).
Practice this: Forks: one piece, two targets →
The plan
Four rules, starting today
- Rook+minor endings. 49% of your 403 endgame mistakes come there. Winning endgames converted: 95%.
- The clock decides the won games. 32 of your 52 losses from +5 ended on time. Median clock when you blunder a winning position: 0:49. At move 30 you have 1:18 in wins, 1:04 in losses.
- Black prep. As Black you are −0.3 by move 10, against +0.4 as White. Start with the Rossolimo and the Najdorf.
- Session length. Over your last 2,000 games you score 82% in the first three games of a session and 71% in games 10–12. After two losses in a row: 72%.
Before every move
- Winning, under a minute: take away their checks.
- Before a capture: what is their most forcing reply?
- Their mistake? Look for the quiet move and the trapped piece.
- Queens off: name the plan once, then play fast.
Week by week
Your own positions, at your own clock
Your game, move 32. You were +5.0. What was the threat?
- Build one file from your games: the 52 losses and 30 draws from +5, plus every position in this report. Run each at depth and mark the move where the eval left +5.
- Daily 15 min: play 5 of those positions out against the engine from the side you had, with the clock you had. All four oversights above came with 37 seconds or less.
- Daily 15 min: two exercises from Aagaard’s Grandmaster Preparation: Calculation. Write the full line, their best defence included, before you check.
- Games: 3+0, rated. After each session, find the first move in each loss where the eval dropped and note your clock at that moment.
Black prep: Rossolimo, Najdorf, offbeat first moves
Move 12 as Black. White’s queen on a7 and bishop on b5 point at your king. What do you need first?
- As Black against the Rossolimo: 14 games, 68%. In the Najdorf: 15 games, −0.29 at move 10, though you still score 80%. Take the lines from your own games through a large database to where they leave theory, then check the critical branches with the engine at depth.
- As White, the Sämisch Nimzo-Indian: 15 games, 60%, your lowest score among your regular White lines. Same method.
- Your offbeat first moves each cost about 8–10 win-% by move 2: 1…f5 after 1.e4 Nc6 2.Nf3 (7 games), 1…b5 against 1.d4 (6), 1.f3 and 1.Nh3 (3 each). If you keep them, prepare the critical reply to each, so the cost stops there.
- Play 20 rated games as Black this week (your Black score: 76%).
Calculation: quiet wins and their forcing replies
Your move as Black, +5.3. White’s queen on d2 and knight on e3 are both on knight squares.
- 172 of your 307 missed punishments were quiet moves; 73 attacked a piece with nowhere to go. Daily 20 min: Aagaard’s Calculation, the chapters on candidate moves and on the opponent’s resources.
- Twice a week: one composed endgame study, solved from the diagram without moving the pieces. The key move is usually quiet, which is the move you miss.
- In your games: before a capture in a sharp position, name their most forcing reply. 98 of the combinations against you started with a check, 95 with a capture sequence.
Converting +5 on a short clock
You have mate, 14 seconds left. Black’s queen and rook are next to your king. Who checks first?
- 32 of your 52 losses from +5 ended on time. The median clock when you blunder a winning position is 0:49.
- Daily 10 min: spar the engine from your own +5 positions (above, and the 52 losses) at 3+0, with the clock you had. The task is to win without letting a check through.
- Winning with under a minute: a forcing finish if you see one; otherwise the move that takes away their checks, even if it is slower on the board.
- You drew 8 winning games by stalemate or insufficient material. When his side is down to a few pieces, count his legal moves before you move.
Rook+minor endings
Your move as Black, +6.0. White’s bishop on g3 is attacked. Take it or keep the rooks?
- 199 of your 403 endgame mistakes came in rook+minor endings, 91 in pure rook endings. Collect the rook+minor positions from your games where the eval dropped and play each out against the engine at the clock you had.
- Dvoretsky’s Endgame Manual, the chapters on rook against minor piece and on rook with minor pieces. Müller and Lamprecht’s Fundamental Chess Endings as the reference when a position from your games is unclear.
- Shereshevsky’s Endgame Strategy for improving pieces before committing. At move 40 your clock reads 0:51 and your median endgame move takes 1.1 seconds: spend the time once, on the move that changes the pawn structure or trades a piece.
Audit again
- Play 60 rated 3+0 games, half of them as Black.
- Re-run this analysis on them. You pass if big mistakes are ≤ 0.4 per 40 moves (now 0.50) and you score 97% from +5 (now 96%).
- If you don’t pass, repeat weeks 1–2 with the new positions.
Games: the latest 2,000 blitz games from the chess.com public API (Mar 23 2025 – Sep 24 2026). Engine: Stockfish 19, 40,000 nodes per position; puzzle lines at 3,000,000. Win% model and accuracy formula: lichess. A big mistake loses ≥ 20 win-% in one move. Rating-point costs: 695 Elo per point of expected score, counting only mistakes the opponent cashed in. Trends: Welch z-test against Prior 9 mo.