9...a5: The Memory Gamble on Move Nine of the King's Indian Defence

On a Thursday night at a small chess club tucked down an alley off Nguyen Thi Minh Khai street in Saigon, I sat behind a twenty-two-year-old student of mine. The blitz clock showed forty-two seconds. On the board was a position I have seen hundreds of times: a white pawn on b4, a black pawn on e5, the pawn chain locked from d5 to e5, both wings yawning open at once. He was playing Black. The knight lifted off f6 and hovered over h5, the classic move of the Mar del Plata variation. I waited for the clock to click.

His hand came back. It hovered seven more seconds, then pushed the a7 pawn to a5.

The next board looked up. At a club where people recite the first fifteen moves of the Mar del Plata the way they recite a multiplication table, 9...a5 is like a man walking into a meeting and asking everyone to put their phones down. Nobody objected. Nobody understood what he intended to do next either.

I wrote the position into my notebook, as I have done for seventeen years. That night I reopened a newly released opening-training package that presents 9...a5 as a remedy against the King's Indian Defence, one the author claims neutralises the Bayonet Attack without any memorisation. What follows is the result of checking that sales pitch against what actually happens on the board.

A system named after a tournament

The King's Indian runs 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6. Black concedes the centre at once, posts the bishop on g7, and announces that the centre will be attacked with pieces before it is occupied with pawns. It is classical hypermodern thinking, born in the 1920s and still entirely current.

The Mar del Plata arises when White pushes the pawn to d5, locks the centre, and Black brings the knight to c6 and pushes e5. The chain freezes from d5 to e5 and turns the board into two separate battlefields. White attacks on the queenside. Black attacks on the kingside. Whoever arrives first wins. The name comes from the 2026 tournament in the Argentine city of the same name, where the variation was played on a scale never seen before and entered the books from there. In the ECO classification it sits roughly between E97 and E99.

What sets the Mar del Plata apart from most openings is the directness of the collision. White may need ten moves to complete a queenside offensive. Black may need exactly ten to push a pawn to g4 and tear open the h-file. There is no room for hesitation. In many elite games the first twenty moves are played almost without thought, because everything already sits in a database.

The Bayonet Attack and why it annoys people

White's ninth move in the main branch is 9.b4, the Bayonet Attack. The idea is blunt: White grabs queenside space, prepares b5 to control c6, prepares c5 to lock the queenside permanently, then brings the bishop to a3, the rook to c1, and the knight round via e1 to d3. White does not need mate. White only needs to freeze Black's queenside before the kingside detonates.

The Bayonet became fashionable in the 1990s and has kept that status since, even if its frequency has declined. The reason is practical: it forces Black to remember. After 9.b4 Black has almost no neutral moves. Every choice leads to a concrete line that can be drilled to move twenty-five with an engine. It is the kind of opening that keeps amateurs awake at night.

Black's orthodox reply is 9...Nh5. The knight hops to the rim, preparing ...f5 and ...Nhf6 and clearing the way for the f-pawn. What follows is a highly forcing sequence: ...f5, ...Nf6, ...c6 or ...c5 depending on circumstances, then ...Kh8, ...Rf7, ...Raf8, ...g5. The whole chain has been analysed to the point where every move carries an engine verdict. That is exactly why it becomes a memory burden for both sides, and the burden weighs heavier on the player with less preparation time.

Which is why 9...a5 appeared as an escape route.

The real mechanism of 9...a5

After 9...a5 the geometry changes. White's b4 pawn, just advanced as a spearhead, instantly becomes a target. Black no longer prepares the kingside first. Black strikes the queenside first, forcing White to answer a question about the pawn before White can ask the opposite question on the other wing.

White has three options, and all three cost something.

First, White takes: 10.bxa5. The a-file opens. Black can recapture with the rook from a8, and though the rook looks exposed, it still controls an open file and pressures a2. White gains the b-file and a little space, but loses the durable pawn structure that is the foundation of the entire Bayonet idea. There is an unwritten rule I have drawn from years of watching queenside play: when you push a pawn forward for space and then trade it on that same wing, you are handing back the gift you just received.

Second, White pushes on: 10.b5. The queenside locks shut. White keeps the structure and controls c6, but also closes the very wing White needs open to attack. With the queenside shut, Black gets a free hand on the kingside with no defensive worries. Every White resource must flow to the kingside, where Black already has more attackers. This is the kind of compromise that leaves many White players feeling steered, even while the evaluation stays level.

Third, White keeps the tension: 10.Ba3, 10.Nd2, 10.Rb1 or 10.Ne1. White continues developing so Black must decide the fate of the a5 and b4 pawns. But waiting in an opening where the opponent has chosen the tempo carries its own risk: Black will calmly complete ...Nd7, ...f5, ...Nf6, ...Kh8, ...Rf7, ...g5, and by then the queenside tension has no value, because the game has already been decided on the other side of the board.

The crux is this: 9...a5 does not create a new position. It relocates the question. Instead of forcing both sides to recall a forcing sequence, it forces both sides to understand a pawn structure. For a player who knows King's Indian motifs, that is an advantage. For a player who is only good at memorising, it becomes a trap pointing the other way.

The trade and what is lost

The promotional material claims 9...a5 leads to sound positions, few concrete lines, and equal chances. The first two claims are broadly correct. The third needs to be read very carefully.

Positional equality does not mean easy play. In the King's Indian, an equal position is usually one where whoever misreads the structure first loses first. White's b4 pawn can be a fulcrum or a fatal weakness. Black's a5 pawn can be a battering ram or a stray piece on the rim. Both possibilities coexist on move ten, and only practical play decides which one it is.

What Black trades away is space. White keeps more squares on the queenside in most branches. What Black receives is freedom of tempo. Black is not obliged to play ...f5 at the moment the books prescribe. Black can wait, play ...Nd7, play ...c6, keep the pawns flexible, and only then open the kingside. That freedom is worth a great deal at club level, where opponents usually win by playing the book correctly rather than by playing well.

What Black loses is the ability to create immediate pressure. After 9...Nh5 White must keep making hard decisions. After 9...a5 White can choose a slow waiting game. In a long time control, letting your opponent wait is no small concession.

The contrarian angle: remedy or commercial product

The summer I spent covering the V-League taught me something: a formation only looks beautiful when the opponent agrees to stand still. The same holds on the board. The move 9...a5 works only if the White player accepts the script Black wants. A well-prepared Bayonet player can choose the branch that keeps tension and push c5 at the right moment, at which point the so-called remedy turns back into an ordinary sideline.

In the database I track, 9...a5 appears at elite level far less often than 9...Nh5. It is a known, long-documented sideline, not a novelty. That does not make it worthless in practice. It does make the word remedy excessive.

There is another detail worth noticing. The material argues that computers often do not know how to handle complex strategic King's Indian positions. That is pre-AlphaZero language. From 2026 onward, neural-network engines evaluate long-term positions far better than the previous generation, and these positions are no longer a blind spot. The claim survives only in a narrow sense: engines assess positions well, but they do not play the next forty moves for a human with a running clock. Saying computers do not understand it is an exaggeration that belongs to an older era.

Esports taught me something football hides well: a 200-millisecond reaction can break an entire system. Chess is the same. An opening branch survives not because it looks good on paper, but because it withstands human decision speed in real conditions. 9...a5 survives at club level because it reduces memory load. Whether it survives at the top is another matter, and only real games answer it.

There is an information-integrity issue I must raise. The package I read carries a title about a team-event round report in Samarkand, while the entire body discusses 9...a5 and the Bayonet Attack. The two are unrelated. The headline may be an aggregation error, or a sign that content was severed from its original context. For readers the consequence is concrete: any information attached to such a product must be verified against the original source before it is cited.

Data limits

This article offers no engine evaluation per branch, no win-draw-loss rates, no average centipawn loss. Those figures are only meaningful with a retrieval date and a database version, and I had neither for this piece.

In 2026 I deleted half of my old dataset, because football after the lockdown was a different sport. The same principle applies to chess in its own way: games before and after neural-network engines became standard should not be mixed in one comparison. A branch written off in 2026 may have been fully re-evaluated by 2026, and the reverse. What should be discarded is engine assessment from the older generation. What survives is pawn structure and kingside attacking motifs, which do not depend on a software version.

What to verify in the next game

No tactic is ever old; only the way we read a game expires.

Three things I will track. Whether 9...a5 shows up in classical games between top players. Whether the frequency of the Bayonet Attack keeps declining, because that measures this product's life cycle. And whether White produces a specific answer to the new branch, as White always does with every opening line that once caused surprise.

My student won that game. What stayed with me is that after it ended, he did not ask which move was correct. He asked why. That is the best sign that he understood what 9...a5 really is: a gamble on memory, not a remedy.

Sources: second-stage technical analysis compiled by the research team, retrieved 12 August 2026; ECO opening classification; reference games from club database. Frequency figures come from a personal tracking sample and do not represent the full international database.