You are up a rook. The board is messy. There are tactics everywhere. Your chess engine shows you a move that keeps all the complications alive — a sharp continuation that requires 15 moves of perfect play to hold the advantage.
This is a bad suggestion.
Not because it is wrong. Because it is unnecessary. When you are winning, the smartest thing you can do is simplify. Trade pieces. Reduce chaos. Convert the advantage without giving your opponent chances to create problems.
Most engines do not think about this. They find the strongest move and stop there. They do not ask: "Does the user need complexity or clarity right now?"
We built a strategy layer that does.
The Problem with "Best Move" in Every Position
Imagine two scenarios:
You are up a queen. The position is relatively quiet. The engine suggests a move that keeps the tension on the board. It is objectively the strongest move. It also gives your opponent the most chances to create counterplay. You blunder on move 37 and lose. You have been there.
You are down a pawn in an endgame. The position is dry. The engine suggests a passive move that slowly loses. No complications, no chances, just a slow death. You resign 20 moves later.
Both are failures of context. The engine found the strongest move in both cases. But the strongest move is not always the most useful move.
Strong human players understand this intuitively. When they are winning, they simplify. When they are losing, they complicate. When they reach an endgame, they activate their king and push pawns aggressively. They change their approach based on the position, not just the evaluation.
We modeled this directly into the engine.
How the Strategy Layer Works
The strategy layer sits on top of the neural network. It does not generate new moves. It re-ranks the candidate moves the network already produced.
The neural network evaluates the position and produces 3 candidate moves with probabilities
The strategy layer reads the game state — material balance, phase of the game, eval trend
It adjusts the ranking based on what the position calls for
The re-ranked moves go to the UI
The network still does the heavy lifting. The strategy layer just nudges the priorities in a direction that makes practical sense.
Winning Positions: Simplify
When the eval shows you are clearly ahead (roughly +1.5 or more), the strategy layer prioritizes moves that reduce risk.
Trade pieces
Every exchange reduces the number of things that can go wrong. A rook endgame up a pawn is easier to convert than a middlegame up a pawn with all pieces on the board.
Avoid unnecessary tactics
If there is a safe move that maintains the advantage and a sharp move that increases it but creates counterplay, the layer picks the safe one. You do not need more advantage. You need to not throw it away.
Reduce tension
When you are winning, tension is your enemy. It gives the losing side chances. The layer favors moves that resolve pawn tensions and clarify the position.
This is not about playing worse. It is about playing smarter. The move might be 0.1 pawns weaker on the evaluation, but it is much harder to blunder.
Losing Positions: Complicate
The opposite applies when you are behind. When the eval shows you are clearly losing (roughly -1.5 or worse), the strategy layer shifts.
Keep pieces on the board
More pieces means more tactical possibilities. More possibilities means more chances for your opponent to make a mistake.
Create imbalances
The layer favors moves that create asymmetric pawn structures, open files, or weaken the opponent's king position. These are positions where the "correct" move is harder to find.
Avoid passive defense
Sitting back and defending usually loses slowly. The layer prefers active counterplay, even if it is objectively slightly worse. A practical chance is better than a slow death.
This is exactly what strong players do when they are losing. They do not accept the slow loss. They create problems. They make the position as messy as possible and hope their opponent cracks under the pressure.
Endgame: Activate and Push
Endgames have their own logic. The strategy layer recognizes when the game has transitioned to an endgame — few pieces, kings become active — and adjusts.
A Concrete Example
Position: You are up a knight. The board still has queens and rooks. The neural network outputs:
Qh5 is sharpest. Strongest objectively. Keeps all pieces on.
Nxe6 trades a knight, simplifies, keeps comfortable advantage.
The player gets a suggestion that helps them convert, not a suggestion that makes the position harder to play.
The Tradeoff
The strategy layer does not find deeper tactics. It does not improve the raw strength of the engine. What it does is make the engine's suggestions more practical for the specific situation you are in.
If you want the absolute strongest move regardless of context, use Stockfish. If you want suggestions that account for whether you are winning, losing, or in an endgame — the way a strong human player would think — the strategy layer is what makes the difference.
It is not about playing better chess. It is about playing the right chess for the position.
See the Strategy Layer in Action
Try Titan Chess and watch how move suggestions adapt when you are winning, losing, or in an endgame.
Read Next
Why Your Chess Engine Should Tilt, Panic, and Get Distracted
Humans tilt, panic under time pressure, and get distracted. Most chess engines ignore this.
Why We Built an Opening Book with 585,000 Positions
Most chess engines do not have an opening book. They calculate from move one. Here is why we built one.