Luxifare · evaluation trace move 10, White to play demo data
Thirty-two concepts scored this position. Four of them produced 61 % of the evaluation's weight, and two of those four argue against the move the engine actually played. The final +0.74 is a net, not a consensus.
FENr1bq1rk1/pp3ppp/2nb1n2/3pp3/8/2NP1NP1/PPP1PPBP/R1BQ1RK1 w - - 2 10
The other 28 concepts — 39 % of the weight between them
Ordered by absolute contribution, largest first, not by concept family. Bars are scaled to the largest single term (118 cp); the four leaders sit above the rule, the remaining 28 below it. Every value is signed: and a + argue for the move, and a − argue against it.
Across 125 tunable parameters in the Rust hot path.
61 % of the total absolute weight, 291 cp of 474 cp.
King safety and pawn structure both say don't.
Down from +1.86 if the two objectors are ignored.
This is the term that decides the move, and it is the one the engine is least able to defend in words. Three sub-terms move at once: shelter quality, attacker count in the king zone, and light-square control around the castled king.
Fires because the g2 bishop carries the light squares around the king on its own, and the term weights that single point of failure heavily. Term weights shelter −29, attackers −27, diagonal −18.
Opening the e-file cuts both ways. The d3 pawn ends up on a half-open file, and the two pawns defending it can only keep doing so by staying where they are.
Fires because c2 and e2 defend d3 only from their starting squares, so the term reads the queenside majority as frozen rather than mobile. Term weights backward pawn −24, half-open file exposure −14.
You are winning a clean pawn, and the e-file opens for whoever can use it first. That is the case for Nxe5, and it is worth about a pawn and a quarter on its own.
You are paying for it with the squares around your own king. The knight that leaves f6 was the piece holding that side together, and g3 does not heal. If you are going to play this, plan on spending a tempo on h3 or Re1 before Black's queen finds the diagonal.
The engine still plays it, at about three quarters of a pawn. Treat the score as a bet you have taken, not a position you have won.
| Concept | Side | cp |
|---|---|---|
| Material + PST | for | +118 |
| King safety | against | −74 |
| Rook activity | for | +61 |
| Pawn structure | against | −38 |
| Mobility | for | +21 |
| Outposts | for | +17 |
| Threats | against | −14 |
| Safe checks | against | −11 |
| Space | for | +9 |
| Bishop vs knight | against | −8 |
| Pawn breaks | for | +6 |
| Endgame technique | for | +4 |
| Conversion | for | +3 |
| Dominance | against | −2 |
| Bishop pair | for | +12 |
| Tempo | for | +7 |
| Open files | for | +6 |
| Piece coordination | for | +4 |
| Candidate passers | for | +3 |
| Connected rooks | for | +2 |
| Blockade | for | +1 |
| Trapped pieces | neither | 0 |
| Pawn storm | neither | 0 |
| Pawn islands | against | −2 |
| Colour complex | against | −3 |
| Queen activity | against | −4 |
| Minor piece placement | against | −5 |
| Weak squares | against | −6 |
| Fianchetto integrity | against | −6 |
| Pawn shelter | against | −8 |
| King zone control | against | −9 |
| Initiative | against | −10 |
| Evaluation after Nxe5 | for | +74 |