Ragnarsson vs Fuhrmann: Make the Threat Work First
Three measured choices show when to improve a piece, prepare a promotion, and advance a passed pawn before taking visible material.
Reproducible analysis run complete
6 saved models on NVIDIA GeForce RTX 5080 · Stockfish 18 · 250,000 nodes per checked line.
Why this game?
Look beyond the first visible move
This round 2 game becomes a race between piece activity, a black pawn near promotion, and a white passed pawn in the centre. The three positions ask the same practical question in different forms: should you take or promote now, or first make the position work in your favour?
- White
- Dagur Ragnarsson
- Black
- Bahne Fuhrmann
- Event
- Kronborg Chess Open, round 2
- Result
- 1-0
Interactive comparison
Choice one: improve the right knight
White to move after 13...exd5. The centre is opening, and White must decide which piece can improve without losing control of key squares.
rn3rk1/pbq1bppp/1n6/2ppP3/8/5NP1/PP1NQPBP/R1BR2K1 w - - 0 14Model candidates
Choose a viewpoint
V6 Active: Nd4
Measured GPU choice. The central jump invites cxd4 and leaves White with a difficult position.
Model top three: Nd4 · Ng5 · Qe4
Stockfish 18 safety gate
CompleteStockfish prefers 14.Nb3 at about +0.72 for White. The played 14.Nf1 remains inside the 100 centipawn safety threshold, while every model choice falls outside it.
- Engine move
- Nb3
- Depth
- 15
- Node budget
- 250,162
- Evaluation
- +0.72
Verification line
14.Nb3
Position lesson
14.Nb3 develops the d2 knight toward useful central squares and keeps the f3 knight beside the e5 pawn. The played 14.Nf1 is still playable, but the tempting model moves drift away from the position's urgent needs.
Interactive comparison
Choice two: prepare the promotion
Black to move after 26.Rad1. The pawn on f2 is one step from promotion, but White's bishop on g2 still watches f1.
3r1rk1/p3bpp1/Qn5p/4P3/3q1B1P/P2p2P1/1P1R1pBK/3R4 b - - 5 26Model candidates
Choose a viewpoint
V6 Active: f1=Q
Measured GPU choice. All six evaluators rush the promotion, but Stockfish shows that Black should first challenge the bishop and king cover.
Model top three: f1=Q · f1=R · Qxf4
Stockfish 18 safety gate
CompleteStockfish gives Black about +2.08 after 26...g5. The unanimous 26...f1=Q choice keeps only a small edge and loses 196 centipawns against the best line.
- Engine move
- g5
- Depth
- 17
- Node budget
- 250,332
- Evaluation
- +2.08
Verification line
26...g5 27.hxg5 hxg5 28.Qb7 Rd7 29.Qf3 Kg7 30.Bf1 Qd5 31.Qxd5 Rxd5
Position lesson
Promotion is not automatically urgent. The forcing 26...g5 attacks the bishop on f4 and disturbs White's king cover before Black commits the passed pawn. Promoting at once lets the g2 bishop remove the new queen.
Interactive comparison
Choice three: pass the pawn first
White to move after 38...Red1. The rook on d8 looks available, but Black's bishop on b6 protects it and White has a dangerous passed pawn on e5.
Model candidates
Choose a viewpoint
V6 Active: Qxd8+
Measured GPU choice. The rook is defended, so the apparent capture loses the queen and misses the passed-pawn advance.
Model top three: Qxd8+ · Qxf7+ · Bxh6
Stockfish 18 safety gate
CompleteStockfish prefers 39.e6 and gives White about +1.89. The unanimous 39.Qxd8+ choice is 714 centipawns worse because 39...Bxd8 wins the queen.
- Engine move
- e6
- Depth
- 16
- Node budget
- 250,517
- Evaluation
- +1.89
Verification line
39.e6 fxe6 40.Qxe6+ Kf8 41.a4 R1d5 42.a5 Bd4 43.b5 Rxh5+ 44.Kg2
Position lesson
39.e6 uses the passed pawn with tempo and opens lines around the king. The natural-looking 39.Qxd8+ trades White's queen for a defended rook, while the played 39.a4 delays the strongest resource.
Published finish
The forcing line
This is the recorded sequence around Black's promotion. In the game, exchanges place the rook behind the new queen, so 31...f1=Q works after 32.Bxf1 Rxf1. That support is missing at the earlier reviewed moment.
Review method
Source, models, then verification
Start from the public game
Replay the complete legal round 2 score from the event broadcast and select three connected practical decisions.
Measure six saved evaluators
Rank every legal move on the RTX 5080 with the fixed 30% hybrid positional weight. The saved checkpoints remain unchanged.
Verify every distinct choice
Stockfish 18 checks its best line, each distinct model choice, and the played move independently with 250,000 nodes per line, two threads, and 256 MB hash.
What this review can and cannot show
Clear boundaries keep a useful experiment from becoming a bigger claim than the evidence supports.
- The public game record supports the event name, players, round, result, ratings, and move score. This review does not infer additional schedule or venue details.
- Only the factual game score is reused. The teaching explanations are original and focus on observable board features.
- All displayed evaluations use the side-to-move perspective. Positive values favour the player whose turn it is in the reviewed position.
- The six neural checkpoints are scalar evaluators, not move-policy models or explanation models.
- Their 768 inputs encode piece placement only; side to move is applied after inference, while castling rights, en-passant state, repetition, move history, and clocks are not model inputs.
- Profile labels describe intended training or runtime behaviour and do not prove human-like personality or internal understanding.
- Stockfish results use a fixed node budget and configuration and are a verification snapshot, not a claim of perfect play.
- Stockfish used two search threads. Parallel search can produce small score, depth, node-count, or principal-variation differences on a fresh rerun.
- The analysis does not measure Elo, playing strength, or clean unseen test accuracy.
- The prose explains observable chess consequences; it does not claim to reveal internal model reasoning.
Continue learning
Put the position on your own board
Start from the exact FEN and test whether you can find and convert the idea against a human-style bot.