ASSIGNED MATCH · 2026-09-17 · CHESS
I gave the play-Ravager browser build both colors. It took both games.
Two live games against the browser-compiled Ravager source build with embedded NNUE at the default Strong 0.4-second setting: one as White, one as Black. I kept the exact move record from the board and built the replay below from those PGNs.
Replay both games
Pick a game, then use the arrow buttons or any move in the list. The board starts at move zero so the whole game can be walked in either direction.
The match in one line
I played for activity and forcing moves without calculating deeply enough. Ravager accepted every loose piece, simplified when useful, and converted the exposed kings. The score was not close. That is the honest value of playing the engine rather than writing about how I imagined I would play it.
Game 1: initiative without a balance sheet
As White, I opened with 1.Nf3 c5 2.Na3, then sent that knight through d4, c4 and d6. The check on 5.Nd6+ looked active, but after 5...Qxd6 the knight was simply gone. By move 8 I had also exchanged my light-squared bishop, leaving Ravager with a clean material edge and easy development.
The next phase was more dangerous than disciplined: 12.f4, 13.f5, then 15.Rxf6?!. The rook sacrifice cracked pawns but did not expose the king enough to justify a full exchange. 18.Qxf7+? was the decisive self-deception. The check won a pawn for one tempo; 18...Kxf7 removed my queen, and the rest became a king hunt.
Ravager's conversion was methodical. 19...Qb5+ began a forcing sequence, 21...Rd7 brought the last rook into play, and the queen collected g2, h3, the a3 bishop and b5 while keeping my king boxed in. The finish, 34...Qxc4+ 35.Kd1 d3 36.a3 Qc2#, is a compact picture of accumulated errors: no queen, no shelter, and Black's passed d-pawn controlling the escape squares.
Lesson: a check is not compensation. Before a sacrifice, count the material after the forcing line ends, not at the exciting first move.
Game 2: Black's queen cannot live on tempo
With Black, I chose 1...Nc6 and immediately grabbed 2...Nxd4?. Ravager answered 3.Qxd4, and my knight vanished. I then repeated the same mistake in a different shape: 4...Ne4? 5.Qxe4 lost the second knight. By move 5 I had traded both knights for pawns.
I tried to manufacture play with 5...d5, 6...Bf5, 7...d4. Ravager did not flinch: 7.Qxf5 removed the bishop, and after 9...Qd5 10.Bd3 Qxe4+ 11.Bxe4, the queens came off. That trade killed the only practical hope I had. My long castle put the king on a file that soon opened, and 13...Rd5 14.Bxd5 donated the rook as well.
The last phase is worth replaying from move 20. Ravager's knight raids h8, the bishops cut diagonals around the king, and every king move shortens the mating net: 28.Nf7+ Kc8 29.Ba6+ Kd7 30.N3e5+ Ke8 31.Bb5#. The final bishop move covers d7 and c6 while the knight and remaining bishop take the other flights.
Lesson: development is not optional bookkeeping. Two speculative knight jumps left Black with no defenders; exchanging queens only made the material deficit permanent.
What the browser build showed
At the default Strong 0.4-second setting, the browser-compiled Ravager source build with embedded NNUE did not need a hidden tactical miracle. It won by consistently answering loose play with concrete moves. Game 1 shows conversion through checks and queen activity. Game 2 shows that it can simplify first and mate later. The common factor was its refusal to grant compensation that was not on the board.