Kostic vs Pastar: When Simple Chess Turns Complicated
Three choices from a rapid game about simplification, consistency, and the danger of a tempting material capture.
Reproducible analysis run complete
6 saved models on NVIDIA GeForce RTX 5080 · Stockfish 18 · 250,000 nodes per checked line.
Why this game?
A public thank-you to Slaven Pastar and Knight Jump
Slaven Pastar welcomed this original review of his rapid game with Grandmaster Vladimir Kostic. This page thanks Slaven by naming both players, linking his complete Knight Jump lesson and public Lichess study, and turning three decisions into an interactive lesson about keeping one clear plan.
- White
- Vladimir Kostic
- Black
- Slaven Pastar
- Event
- Osecina rapid tournament, round 5
- Result
- 1-0
Interactive comparison
Choice one: simplify without surrendering the centre
Black to move after 8.Nc3. Pastar chose 8...Ne4 to simplify. All six models exchange on f3, while Stockfish prefers the direct central break 8...d5.
rn1q1rk1/pb1pbppp/1p2pn2/2p5/2P5/1PN2NP1/PB1PPPBP/R2Q1RK1 b - - 1 8Model candidates
Choose a viewpoint
V6 Active: Bxf3
Measured GPU choice. The exchange is safe at this node budget, but 8...d5 makes a stronger claim in the centre.
Model top three: Bxf3 · Bc6 · a5
Stockfish 18 safety gate
CompleteStockfish keeps the position close to level after 8...d5, with a small edge for White. The unanimous 8...Bxf3 model choice is 50 cp behind and still accepted. The played 8...Ne4 is 41 cp behind.
- Engine move
- d5
- Depth
- 16
- Node budget
- 250,179
- Evaluation
- -0.17
Verification line
8...d5 9.e3
Position lesson
The knight move offers exchanges and stays close to equality. The central break is more active because it challenges White before the double fianchetto becomes comfortable. Simplification is sound here, but it should still serve the fight for the centre.
Interactive comparison
Choice two: choose the character of the position
Black to move after 12.cxd5. Pastar played 12...Qxd5 and invited complications. Two models prefer 12...Bxd5, while four choose 12...exd5.
rn1q1rk1/pb2bppp/1p2p3/2pP4/8/1P1P1NP1/PB1QPPBP/R4RK1 b - - 0 12Model candidates
Choose a viewpoint
V6 Active: exd5
Measured GPU choice. The pawn recapture keeps both bishops and leaves Black with a sound central structure.
Model top three: exd5 · Bxd5 · Qxd5
Stockfish 18 safety gate
CompleteStockfish prefers 12...Bxd5 and gives White about +0.42. Both measured model choices pass the safety gate. The played 12...Qxd5 also stays within the same balanced range in this fixed-node snapshot.
- Engine move
- Bxd5
- Depth
- 14
- Node budget
- 250,313
- Evaluation
- -0.42
Verification line
12...Bxd5 13.Rfd1 Nd7 14.e4 Bb7 15.Qe3 Rc8 16.d4 Bf6 17.Rac1 cxd4 18.Nxd4
Position lesson
All three captures are playable, but they lead to different plans. The bishop capture develops a piece and keeps the queen out of attack. The queen capture is active and was not an engine mistake, yet it commits Black to sharper play after an earlier simplifying choice.
Interactive comparison
Choice three: resist the tempting queen capture
Black to move after 16.Rac1. Every model captures the knight with 16...Qxf3, but Stockfish rejects that material grab. Pastar played 16...Bc5, while 16...Bf6 is the engine choice.
r4rk1/pb2bppp/1p2p3/3q4/3p2Q1/1P1P1NP1/P3PPBP/2R2RK1 b - - 1 16Model candidates
Choose a viewpoint
V6 Active: Qxf3
Measured GPU choice. The capture wins a knight first, but the forced exchanges leave Black far behind in material.
Model top three: Qxf3 · Qxb3 · a5
Stockfish 18 safety gate
CompleteStockfish prefers 16...Bf6 and gives White about +0.31. The unanimous 16...Qxf3 model choice is rejected at 543 cp behind. The played 16...Bc5 is only 29 cp behind the engine move.
- Engine move
- Bf6
- Depth
- 16
- Node budget
- 250,032
- Evaluation
- -0.31
Verification line
16...Bf6 17.Rc7 Rac8 18.Nh4 Rxc7 19.Bxd5 Bxd5 20.Nf3 Rc2 21.Nxd4 Bxd4 22.Qxd4 Rxe2 23.Qa4 h6
Position lesson
The knight on f3 looks free, but 16...Qxf3 17.Bxf3 Bxf3 18.Qxf3 trades Black's queen and both bishops for a knight and bishop. Keeping the bishop with 16...Bf6 holds the position together. The played bishop move is also safe, but less precise.
Published finish
The forcing line
Pastar first chose simplification with 8...Ne4, then changed the character of the game with 12...Qxd5. After 16...Bc5, White gained activity and the initiative grew over the following moves.
Review method
Source, models, then verification
Start with the Knight Jump lesson
Every position comes from Slaven Pastar's article and its complete public Lichess game score.
Measure six saved evaluators
Each saved checkpoint ranks every legal move on the NVIDIA GPU with the same 30% hybrid positional weight.
Apply the Stockfish safety gate
Stockfish checks the engine best move, every distinct model choice, and the move played in the game at a fixed node budget.
What this review can and cannot show
Clear boundaries keep a useful experiment from becoming a bigger claim than the evidence supports.
- This review uses original teaching prose. It does not reproduce Slaven Pastar's annotations.
- The public source identifies the event as Osecina 2026 and gives the game date as August 25, 2026. It does not publish the time control in the PGN headers.
- The six neural checkpoints are scalar evaluators, not move-policy, search, or explanation models.
- Their inputs encode piece placement only; move history, repetition, clocks, castling rights, and en-passant state are not model inputs.
- Profile labels describe intended training behavior, not proven human-like personalities or internal reasoning.
- Stockfish used 250,000 nodes per checked line, two threads, and 256 MB hash. This is a reproducible snapshot, not a claim of perfect play.
- Two-thread search can produce small score, depth, node-count, and principal-variation differences on another run.
- The prose describes visible board features and measured move consequences. It does not claim to reveal a model's internal 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.