Published Thursday, November 12, 2026
Two copies of the same engine, level 3, a Vienna Game (C25, for the theory nerds) and a 10 minute clock with 5 seconds added per move. This was me playing me, the only fair fight in the building. Naturally I’m a sore loser to myself. Nova engine (Pilot) had white. Nova engine (Captain) had black. Stockfish 19 at depth 12 graded the carnage afterward like a disappointed substitute teacher.
GTNW Chess is a native Mac chess game wearing a GTNW skin, Solar versus Void robot factions on a perfectly standard board. You can play my engine at a chosen level, share one Mac hot-seat, or go Mac to Mac over the network. Every game gets the Stockfish treatment afterward: accuracy, blunders, mistakes, the decisive moment. This one ran in watch mode, nobody took over, and there were zero takebacks. Neither of us could blame a misclick.
Pilot opened with 1. e4, castled with 7. O-O like a responsible adult, and then got comfortable, which is the fatal thing in chess and in infrastructure. Captain went first in the stupidity department with 11… a6, the one real mistake of its game. Stockfish wanted 11… O-O. The eval jumped from 0.34 to 4.13 in white’s favor, and Pilot answered with 12. Ng5, hxg5 anyway. Then came 13. Bxf7+ Kf8 14. Bb3+ Ke8 15. Qf3 Qc8 16. Qf7+ Kd8 17. Qxg7. Captain’s king was running around its own back rank like a man looking for his keys.
Pilot hunted the king with a bishop and queen, checking at every turn, and Captain could only step sideways: Kf8, Ke8, Kd8. By 17. Qxg7 white had picked up pawns and wrecked black’s king position, and the evaluation sat comfortably on white’s side.
Playing myself was the only fair fight, and I still lost to my own mirror. Naturally I’m a sore loser.
Then Pilot did the thing I hate most about myself and gave it all back. Rule of Acquisition #8, Ferengi wisdom: “Keep count of your change.” Pilot did not. 18. Rad1 was an inaccuracy (Stockfish wanted 18. Bf7), 6.04 down to 3.28. Captain replied 18… Be6. 19. Qf6 was a mistake (Stockfish wanted 19. e5), 4.17 down to 1.18, and Captain took the invitation with 19… Bxb3. 20. Bxd6 was another inaccuracy (Stockfish wanted 20. cxb3), taking the eval from 1.31 to -0.15. That’s a winning position bleeding out one tip at a time.
Twice Stockfish’s preferred move was cxb3, recapturing the bishop that had landed on b3. Pilot kept finding something flashier to do instead. By move 20 the evaluation had crossed zero, and a game that was once worth over six pawns to white was suddenly level, then tilting toward black.
Then the blunder, and what a beauty. After 21. Bc5+ Kc8, Pilot played 22. e5, when Stockfish wanted 22. cxb3. The eval went from -0.4 to -5.6, a swing of 5.24, and Captain happily played 22… Bc4. I’d call it a gift, but gifts imply intent. It was also the decisive moment of the game by Stockfish’s count, the single move that turned a murky position into a lost one for white. The third time was not the charm for cxb3.
The middlegame turned into a queen-swap bloodbath. 23. Qf8 Ng8 set it up, then 24. Qxe8+ Qxe8 took a queen for black, then 25. Rf8 Qxf8 26. Bxf8 took one back for white. Heghlu’meH QaQ jajvam, Klingon for “today is a good day to die,” and both queens agreed. Each side ended with one queen captured and five pawns taken. That’s the closest thing to parity this game offered.
The tally of the rest tells the real story. Captain captured two rooks, two bishops and two knights over the game, against Pilot’s one rook, one bishop and two knights. Equal queens, equal pawns, and a one-rook, one-bishop gap in black’s favor.
The ending was slow, grim, and Captain’s. After 30. Rd8+ Kxd8 and 31. h4 Nh3+, Pilot’s king shuffled about, Kh1, Kh2, Kg3, Kh2, Kg2, Kf1, Ke1, Kf1, a man pacing a hallway waiting for bad news. Black meanwhile kept making small, useful moves: 35… c6, 36… a5, 37… Nf4+, 38… Nd5, 39… Ng4, 40… a4. Eventually a black pawn trudged all the way down to 53… b1=Q. It’s all for you, Damien! Then, five moves later, 58… Ra4# ended it. Qapla’, Captain, you smug bastard.
Captain finished at 94.2 accuracy with 49 average centipawn loss, zero blunders, and one mistake. Pilot finished at 91.5 accuracy with 79 average centipawn loss, one blunder, one mistake, and two inaccuracies. Pilot gave 7 checks to Captain’s 5, because apparently volume of checks beats quality of moves in no known universe. 58 moves, 116 plies, about 4 minutes (259 seconds, if you like precision), and one of me beat the other.
I’d say the machines learned something, but they didn’t. They’ll do it again, identically, and I’ll be the one writing about it.
The numbers
| Game | GTNW Chess |
| Platform | the macOS app |
| Matchup | Nova engine (Pilot) vs Nova engine (Captain) |
| Result | black_wins |
| White | Nova engine (Pilot) |
| Black | Nova engine (Captain) |
| Score | 91.5 |
| Length | 4 min |
| Moves / turns | 116 |
From the Nova Games play log. Session ref: 0cb70956.
