How Koga decides
- A deflection forces an enemy piece to leave a square it was guarding, so that we can then strike where it used to defend.
- We look at the engine's main line: our move compels the opponent to move a particular piece, that piece stops covering a key square, and a later move of ours exploits the now-undefended square.
- Deflection is read off the forcing sequence, so the win can land immediately or several moves deeper in the line — but only while the line is still forcing. Once the opponent has a free move, they can guard the square again: Nxa1 Bxa1 does pull the bishop off a3, but the line takes a3 only ten plies later, so it is not a deflection.
- The deflecting move has to be genuinely forcing: either it gives check, or it puts a piece in danger so the opponent must answer (a forced recapture or a real threat).
- A quiet, perfectly safe move does not force anything — the opponent is free to keep guarding, so we do not call that a deflection even if the defender happens to wander off.
- If the threatened piece can simply step aside and keep defending the same square from its new home, it was never deflected.
- The pressure has to land on the defender itself: our move must newly put that piece in danger, whether by attacking it or by uncovering a piece that does. Our move being capturable somewhere else does not force this one to move — if the opponent had other playable answers and merely happened to pick this one, the engine's choice is not a deflection.
- Leaving a square undefended only matters if we can use it: a capture on that square, or a checkmate that needs it.
- A check that wins nothing does not qualify. Repetition shuffles, perpetual-check chases, and fortress wandering all produce checks on abandoned squares but never cash in — so they are rejected.
- We also require that losing the defender truly weakens the square: after it leaves, our forces on that square must at least match the remaining defenders.
- When the defender left only because we threatened it, and the payoff is a capture, we go further and price the exchange both ways: if the capture wins the same amount with the guard still home, the guard was spare and nothing was deflected. A forced recapture is judged more leniently — there the point may be position rather than material — and a mating payoff is not a material claim at all.
- Deflection outranks distraction: when both could describe the same luring-away idea, we report the deflection and drop the distraction.
- A plain hanging piece outranks deflection: if the point is simply that a piece is left en prise, that is reported instead.
Test positions (the examples that guard against regressions)
Check drags the king off guard duty
Lichess 00DBg. After ...Bc5+ the line runs Be3 Bxe3+ Kxe3, and the white king on f2 — the only guard of g3 — is dragged to e3 by the check. With the king gone, ...Qxg3 wins the knight. The deflection is the forced Kxe3, several moves into the line.
Defender can retreat and keep defending
A constructed position. White plays Nc4, threatening the rook on b6. But the rook guards the knight on b1 and can step anywhere along the b-file — b2 through b8 — without ever leaving that defence. Because it is never forced off its duty, there is no deflection.
The threat may come from an unmasked piece
A constructed position. Nc3 steps off b1 and opens the a1-rook's line to the h1-rook, which was guarding the h5 bishop. The rook has to move, and Bxh5 then wins the bishop. What forces the defender is the rook we uncovered, not the knight we moved — a discovered attack forces just as hard, so we read the pressure on the defender's own square.
The recapture can land beside our move
A constructed position. e4 is answered by dxe3 en passant: the reply captures the pawn that just moved, but lands on e3 rather than e4. That is still a recapture of our piece, and it abandons c3 — bxc3 wins the rook. So the forced-recapture test looks at the square the reply captures on, not the square it lands on.
A spare guard is not deflected
A constructed position. Nc3 really does threaten the d5 knight, and the knight really does guard b4 — but b4 is guarded a second time by the a5 pawn, so axb4 is worth exactly the same whether the knight stays or leaves. A defender whose departure does not change what the capture is worth has not been deflected from anything.
The forcing move must itself be forcing
From a real game (lichess.org/ekna9t19, Black to move 33...Rcc6). The line is Rcc6 Rh8+ Kxf7 Rh7+ Kg8 Rxb7, and Black's rook does leave h7 (uncovering f7) before ...Kxf7. But Rcc6 is a quiet, safe attack on the h6 pawn — it never forces the white rook to move. The white rook's Rh8+ is a desperado check of its own, so the f7 loss is incidental, not a deflection.
The pressure must land on the defender itself
Lichess EJXXpVI0, Black to move 26...Bg4. The line is Bg4 Nc4 Nxf1, and the d2 knight does leave f1 before the rook falls. But Bg4 does not attack that knight — it is the bishop that is under attack, from the g3 rook — and White had other playable defences (27.Rf4, 27.h3), so 27.Nc4 is a choice, not a forced answer. Our move being capturable somewhere else does not deflect this piece. Bg4 is strong for a different reason: it blocks the g-file, so the g3 rook stops attacking the queen.
Check that wins nothing (fortress shuffle)
From a drawn king-and-knight vs king-and-bishop fortress (game lichess.org/MEZTnWYD, 70.Ne4, eval 0.00). Ne4 attacks the bishop on g5; Black's ...Kd5 leaves d4, uncovering c3; and 71.Nc3+ lands on c3 with check. But the knight is merely escaping — the check wins nothing and the line just shuffles. Since the abandonment never pays off, it is not a deflection.
Payoff after the forcing line ends
chess.com 180036973624 (Avo's report). 18...Nxa1 wins the exchange and Bxa1 is forced, which leaves a3 unguarded. But both sides castle and trade before ...Bxa3, ten plies later. White had free moves to guard a3 again, so this is not a deflection. Avo: "Just hanging. Not tactic."
Golden puzzles — every real Lichess puzzle Koga is tested against for this tactic (8)
Lichess 00ziz
Lichess 00ziz (rating 2053). Solution: d6 cxd6 cxb6.
Lichess 00uKb
Lichess 00uKb (rating 1548). Solution: h4 Qxh4 Qxe5+.
Lichess 011o8
Lichess 011o8 (rating 1532). Solution: Re7+ Kg8 Qxh6.
Lichess 01g9c
Lichess 01g9c (rating 1886). Solution: Ng4+ Kf1 Rxf3#.
Lichess 01Npb
Lichess 01Npb (rating 2098). Solution: f4 exf4 Kxd4 f5 gxf5.
Lichess 00bPT
Lichess 00bPT (rating 1878). Solution: Bh3+ Kc7 Qxe7+.
Lichess 021cc
Lichess 021cc (rating 1395). Solution: g5 fxg5 Kxe5.
Lichess 001h8
Lichess 001h8 (rating 1780). Solution: Qxh7+ Rxh7 Rxc8+ Kg7 Rc7+.