Experiments5 October 202610 min read

I won this game. Stockfish says I was losing for twenty moves.

The last post was a four-move loss. This one is the opposite: a 61-move game I won, played on 1 October as Black, rated 582, against a 545. It was a 3+2 blitz game and it ended in checkmate. I fed it to the same two coaches and checked what they said the same way, by replaying every concrete claim in chess.js.

Before the outputs, one fact that changes how you read them. Stockfish thinks I was losing for about twenty moves of this game. I only won because White hung their queen.

The game

Show the full PGN (61 moves)

[Date "2026.10.01"]
[WhiteElo "545"]
[BlackElo "582"]
[TimeControl "180+2"]
[Result "0-1"]
[Termination "Black won by checkmate"]

1. e4 c5 2. Nf3 Nc6 3. d3 d5 4. h3 g6 5. Nbd2 Bg7 6. c3 Nf6 7. e5 Nd7 8. d4 cxd4
9. cxd4 e6 10. Nb3 O-O 11. Be2 Nb6 12. Bg5 Qc7 13. O-O Nc4 14. Qc2 Bd7 15. Nc5
Nb6 16. b3 Nb4 17. Qc3 Na6 18. a3 Nxc5 19. dxc5 Nc8 20. b4 b5 21. Bf4 Bc6 22.
Qd4 f6 23. h4 fxe5 24. Nxe5 Bxe5 25. Bxe5 Qf7 26. f3 Ne7 27. Bd3 Nf5 28. Bxf5
exf5 29. Kf2 Rae8 30. Rh1 Qb7 31. Rae1 Re7 32. h5 Rfe8 33. hxg6 hxg6 34. Qh4 Rh7
35. Qg5 Qf7 36. Rxh7 Qxh7 37. Bf6 Rxe1 38. Kxe1 d4 39. Kd2 Qd7 40. Qxg6+ Kf8 41.
Bg7+ Qxg7 42. Qxc6 Qxg2+ 43. Kd3 Qg1 44. Qf6+ Ke8 45. Qxf5 Qd1+ 46. Ke4 Qb1+ 47.
Kxd4 Qxf5 48. Ke3 Qe5+ 49. Kf2 Qb2+ 50. Kg3 Qxa3 51. Kg4 Qxb4+ 52. f4 Qxc5 53.
f5 Qf8 54. Kg5 b4 55. Kf4 b3 56. Kg5 b2 57. Kg6 b1=Q 58. Kg5 Qbxf5+ 59. Kh4 Qh6+
60. Kg3 Qhg5+ 61. Kh2 Qh7# 0-1

The engine data EloCoach produced for this game flags four turning points. Numbers are White’s winning chances, before and after the move.

Move	Engine’s best	White’s chances	Who
15...Nb6	Nxd4	36% to 76%	me
24...Bxe5	Rxf4	23% to 86%	me
44...Ke8	Kg8	52% to 85%	me
45.Qxf5	Qe5+	85% to 4%	White

From 24...Bxe5 until 45.Qxf5, White stayed ahead the whole way, between about 52 and 97 percent. After 45.Qxf5 it was 4 percent and the rest was technique. So a fair review of this game has to say something like “you were in trouble for most of it and got a gift”. Keep that in mind.

What Gemini did

This time Gemini was not working from the raw text alone. It ran some Python on my PGN first, replayed all 122 plies and printed every position, then wrote its review. So it had a real board to look at. That is a fairer baseline than the last post, and I want you to know about it.

Read Gemini’s full review

The Python it ran is left out. Its text is otherwise as written, with two dashes replaced by hyphens.

Great game! You pulled off a clean victory with Black. At the ~580 Elo level, staying alert for tactical opportunities and executing endgame wins cleanly is what drives real rating growth - and you did both here.

Here is a breakdown of your play, highlighting your winning tactical sequence and key areas to refine.

The Game-Winning Turn: Move 46... Qb1+!

The star moment of the game came right after White pushed their king into the center:

The Setup (Move 42-45): After a chaotic piece trade, White picked up your bishop on c6 with 42. Qxc6. You counter-attacked White's pawns with 42... Qxg2+ and kept up the pressure.
The Tactic (Move 46): White played 46. Ke4, stepping directly onto the diagonal between your queen on d1 and their queen on f5.
The Execution: You immediately spotted 46... Qb1+! - a beautiful skewer! White was forced to move their king (47. Kxd4), leaving their queen completely undefended. You snapped it up with 47... Qxf5, effectively sealing the game.

What You Did Well

Lethal Tactical Vision: Spotting the 46... Qb1+ skewer under time pressure (180+2 blitz) was sharp. Many players at this level miss king/queen alignments on diagonals.
Flawless Conversion & Mop-Up: After winning the queen, you didn't hesitate:
You methodically swept White's passed pawns (49... Qb2+, 50... Qxa3, 51... Qxb4+, 52... Qxc5).
You pushed your passed b-pawn directly to promotion (54... b4 $\rightarrow$ 57... b1=Q).
Clean Checkmate: You created a second queen and delivered a crisp ladder mate with 61... Qh7#, giving White zero stalemating tricks or escape squares.

Key Areas to Improve

1. Watch Out for Undefended Pieces ("Loose Pieces Drop Off")

Move 41-42: When White played 41. Bg7+, you captured with 41... Qxg7. However, your bishop on c6 was left undefended, allowing White to play 42. Qxc6 for free.
Takeaway: Before executing a forced recapture, take a split-second scan of the rest of the board to check if any of your other pieces suddenly lose their defender.

2. Avoid Getting Knights Trapped on the Back Rank

Moves 13-19: You maneuvered your knights back and forth (...Nc4, ...Nb6, ...Nb4, ...Na6), which allowed White to expand with b3, a3, and dxc5. This eventually forced your knight all the way back to c8.
Takeaway: Aim to place knights on stable outposts (like c5 or d5) where enemy pawns cannot easily kick them away.

3. King Safety Against h-File Attacks

Moves 32-36: White opened the h-file with 32. h5 and 33. hxg6, bringing their queen and rooks dangerously close to your king (34. Qh4). You defended well with 34... Rh7, but allowing an opponent to rip open the h-file in front of a castled king is high-risk.

Quick Summary

Phase	Grade	Highlights
Opening	B+	Solid Sicilian structure (`c5`, `d5`, `g6` Dragon setup).
Middlegame	B-	Got pushed back slightly; lost `c6` bishop, but defended the h-file attack.
Tactics	A+	Spotted `46... Qb1+` skewer to win White's queen.
Endgame	A+	Clean pawn cleanup, promotion, and 2-queen checkmate.

It is readable and encouraging, and the tactical story is right. 46...Qb1+ is a skewer, White’s king did have to move, and the queen on f5 fell. Here is what I could not make hold up:

“Stepping directly onto the diagonal between your queen on d1 and their queen on f5.” Those two squares are not on the same line. The skewer comes after Qb1+.
Its first improvement tip says 41...Qxg7 left the bishop on c6 loose and White took it “for free”. It was not free. I had just taken White’s bishop on g7, so it was an even trade, and the engine data rates 41...Qxg7 as a good move. The lesson is built on a position that was fine.
“Under time pressure.” The PGN has no clock times, so it cannot know that.
The grades: opening B+, middlegame B-, “got pushed back slightly”. Stockfish had me clearly worse for twenty moves. Gemini’s review reads like a comfortable win.

It also never mentions the three moves the engine marks as my biggest mistakes: 15...Nb6, 24...Bxe5, 44...Ke8. It does say my knights shuffled around moves 13 to 19, which is the right neighbourhood for 15...Nb6, so I will give it that.

What EloCoach did

Read EloCoach’s full review

Taken from the database record. The app’s board diagrams and section tags are removed, each link to one of my tracked habits shows as “[habit link]”, and a dash is replaced by a hyphen. The words are otherwise what the coach wrote.

The Opening: Sicilian Defense, Old Sicilian Variation (3.d3)

You answered 1.e4 with the Sicilian, and White steered into the quiet waters of the Old Sicilian with 3.d3. Against an early d3, Black usually seizes immediate space in the centre with d5, which you played on move 3. White adopted a King's Indian Attack setup with Nbd2 and an early h3, while you developed harmoniously with a kingside fianchetto.

The structure locked on move 7 when White pushed e5 and followed with d4. You traded on d4 with cxd4 on move 8, leaving White with an isolated-looking pawn wedge on d4 and e5, while your dark-squared bishop on g7 bit on granite. Everything remained balanced and solid through castling.

The Middlegame: Complications and Tactical Crossfire

The battle ignited on the queenside. On move 13, you planted your knight forward with Nc4, prompting White to step their queen to c2. White then lunged forward with 15.Nc5?, offering an exchange while leaving their d4-pawn unguarded.

The Missed Blow on d4

When White jumped into c5, their queen on c2 was tied to defending the knight, and the pawn on d4 was defended only by the knight on b3. You had an immediate tactical strike available with 15...Nxd4!, which wins a central pawn and unravels White's queenside. Instead, you retreated your other knight with 15...Nb6??, missing the tactic and handing White a heavy positional grip on the c-file.

White pressed forward, but gave you another chance with 18.a3?, hitting your knight on b4. Here, you found the sharp resource 18...Nxc5!!.

Rather than meekly retreating the attacked b4-knight, you recognized that taking on c5 breaks White's queenside clamp and forces White to capture on c5 with 19.dxc5, leaving their central pawn structure split. Over a quarter of players at your level spot this capture; you saw the key idea that strong players see.

The Pin and the Free Rook

The tension shifted back to the centre. On move 22, you pushed 22...f6??, trying to chip away at White's e5-pawn, but opening dangerous lines toward your own king. White missed 23.exf6 and instead played 23.h4??, allowing your great reply 23...fxe5!.

White then made a fatal miscalculation with 24.Nxe5??, walking their knight directly into a pin.

White's bishop on f4 was completely undefended on an open file. Your rook on f8 attacked the f4-bishop directly, shielded only by the knight on f5. Taking the knight on e5 with 24...Bxe5?? was an exchange that gave White an escape hatch.

You should have played 24...Rxf4!, which wins the undefended dark-squared bishop outright because White's e5-knight cannot recapture while pinned to the white queen. After your capture on e5, White recaptured with 25.Bxe5, keeping their dangerous bishop alive and turning their sights toward your kingside.

The Turning Point: The Endgame Scramble and Queen Snare

White mounted a heavy attack down the h-file, forcing heavy piece trades. You held your nerve under heavy fire, exchanging rooks with 37...Rxe1!! on move 37 and simplifying the position.

White won a pawn on g6 with check, and after your accurate 41...Qxg7!!, the queens came off the dark-squared diagonals into an open queen-and-pawn endgame.

Surviving 44...Ke8 and the Game-Winning Skewer

The game hung by a thread on move 44. Under check from White's queen, you played 44...Ke8??. This stepped your king into the line of fire, where White had 45.Qe5+ with a crushing attack. Most players at your level find 44...Kg8 to tuck the king into safety.

White, however, got greedy and snapped off your f-pawn with 45.Qxf5??, completely missing the geometry of the board.

You immediately punished the blunder. You checked White's king with 45...Qd1+, driving White's monarch to e4, and then landed the decisive blow: 46...Qb1+!.

This delivered a deadly skewer. White's king stood on e4 on the b1-h7 diagonal, directly in front of the unprotected white queen on f5. White had to step out of check with 47.Kxd4, leaving their queen completely undefended. You snapped it up with 47...Qxf5!, sealing the game on the spot.

Conversion: The Outside Passed Pawn

Once White's queen fell, you gave White no hope of counterplay. You pushed White's king back with checks, devoured White's remaining pawns on a3 and c5, and created an unstoppable outside passed pawn on the b-file.

You marched the pawn straight down the board: 54...b4, 55...b3, 56...b2, and 57...b1=Q, promoting to a second queen. With two queens dominating the board, you boxed White's king into the corner with 60...Qhg5+ and finished the battle with 61...Qh7#.

This game showed the exact tactical sharp eye that defines your play [habit link], spotting the royal skewer that won the queen, but it also exposed your recurring blind spot: missing undefended pieces when responding to threats [habit link].

Strengths

Calculated liquidation: Your choice of 18...Nxc5!! broke White's queenside clamp and demonstrated real composure under pressure when your pieces were being pushed around.
Flawless endgame conversion: Once you won White's queen on move 47, your technique was clinical [habit link]. You created an outside passed pawn, escorted it straight to b1, and mated without allowing a single check.

Areas to work on

Before you move, find every undefended piece, yours and theirs: On move 24, White jumped their knight into e5. You automatically traded pieces with 24...Bxe5??, focusing only on the knight that just moved. If you had scanned the whole board, you would have seen that White's bishop on f4 had no defender whatsoever, and 24...Rxf4! wins the piece immediately.
King escape paths under queen checks: On move 44, choosing 44...Ke8?? instead of 44...Kg8 stepped your king into the open center where White had crushing checks. In queen endgames, prioritize hiding your king behind remaining pawns rather than running into open files.

Coach's verdict

This game was decided the second White grabbed your f5-pawn and walked into your queen skewer on b1. You showed grit by defending an uncomfortable h-file assault, and when the window opened, your opportunistic tactical eye didn't miss [habit link].

To climb past the 600 mark and bring your blitz rating closer to your 1000 rapid baseline, you must stop reacting solely to the piece that just moved and start scanning for loose enemy pieces across the entire board.

On move 24, when White played 24.Nxe5, what made you play 24...Bxe5? Were you focused entirely on eliminating the jumping knight, or did you consider capturing the bishop on f4 with your rook?

I only looked at taking the knight

I saw the bishop but didn't realize it was undefended

The big difference is where it looks. It goes straight to the three moves the engine flagged and names the better one each time: 15...Nxd4, 24...Rxf4, 44...Kg8. All three are the engine’s best moves. It also calls out White’s 23.h4 and 45.Qxf5 as the errors that let me back in, again matching the engine. Gemini did not find any of this.

Then it explains why those moves work, and that is where it falls apart. Replaying the positions in chess.js:

On move 15 it says White’s d4 pawn was “defended only by the knight on b3”. That knight had just moved to c5. The pawn is defended by the knight on f3.
It says White’s queen on c2 was “tied to defending the knight” on c5. It was not. The knight is defended by the d4 pawn only, and Black’s knight on c4 blocks the queen anyway.
On move 24 it says White’s bishop on f4 was “completely undefended” and had “no defender whatsoever”. The queen on d4 defends it.
It says my rook was “shielded only by the knight on f5”. I had no knight on f5 at that point. My only knight was on c8.
It says 24...Rxf4 “wins the bishop outright” because nothing can recapture. Qxf4 recaptures. The move is still the engine’s best, but the reason given is wrong.

There is a real explanation for 15...Nxd4, and it is not the one in the review. The knight on c5 is defended only by the d4 pawn. Taking that pawn removes the defender, and moving my c6 knight opens the c-file so my queen attacks c5. The engine tags it exactly like that: removing the defender, discovered attack. The coach had the right idea and the wrong geometry.

The other thing is that the coach’s lesson of the game, “find every undefended piece”, is half off target. Rxf4 works because of a pin and a clearance, not because the bishop was loose. It was defended.

The “brilliant” problem

The review marks 18...Nxc5, 37...Rxe1 and 41...Qxg7 with “!!”, and the app’s engine data calls all three “brilliant”. Each one is an even trade. 18...Nxc5 dxc5 is knight for knight. 37...Rxe1 Kxe1 is rook for rook. 41...Qxg7 Qxc6 is bishop for bishop. They are fine moves, but “brilliant” oversells them, and that looks like a bug in how the app decides what counts as a sacrifice. I wrote up how that label is built, and how it has gone wrong before, here.

The scoreboard, as honestly as I can do it

What I checked	Gemini	EloCoach
Finds the engine’s three biggest mistakes of mine	None by name	All three, with the better move
Says I was worse for twenty moves	No, reads as a comfortable win	Only implied
Board facts I could prove false	2, plus 1 it could not know	5, all in the explanations
Lesson built on a wrong position	Yes (41...Qxg7)	Partly (the “undefended” label)
Ties the game to my earlier games	No	Yes, from stored notes I cannot see here
Asks me a question, offers next steps	No	Yes
Over-generous labels	None I found	Three “brilliant” trades

I did not check whether the habit link in EloCoach’s review is accurate, since I cannot see the notes behind it. And I only counted statements about the board that chess.js can prove or disprove. Opinions like “this was a lapse in concentration” are not in these numbers.

What I take from it

The engine decides what to talk about, and it matters a lot. Gemini had a board but no evaluation, so it missed where the game was actually lost and saved. EloCoach looked at the right three moments because Stockfish told it to.
The language model still writes the “why”, and that part is shaky. Five wrong statements about who defends what, all in the sentences that explain the tactic. It is the same kind of mistake as in the four-move game.
So this is not a clean win. On this game EloCoach found more and said more, and was wrong in more places. I would rather tell you that than pick a number that flatters us.

The fix I am considering is the one from last time, and this game would be a clean test of it. For each key move, work out in code what is defended by what and what each move attacks, and hand that to the coach as a table. If the table had said “d4: defended by Nf3” and “Bf4: defended by Qd4”, most of the five errors above could not have been written. I have not decided to build it.