9...a5 and the Battle Around the Bayonet Attack: Rereading a King's Indian Theory Package

Saturday morning, table three of a chess club tucked down an alley off Nguyen Thi Minh Khai Street. White has just pushed the b2-b4 pawn into the queenside space, and all four people around the board know the next move: Black's knight hops from f6 to h5, opening the familiar race. The man playing Black, an engineer in his forties, does not touch the knight. He pushes the a7-a5 pawn. The table goes quiet for a few seconds. White, who had memorised the first fifteen moves of the main line, suddenly has to think for himself.

I stayed until that game ended and wrote the whole sequence into my notebook. That 9...a5 is something I rarely see at amateur boards in Saigon, where people walk the road they were taught. Over the past two months it has resurfaced in opening-theory packages sold on chess content markets: a recorded lecture product promoting 9...a5 as an antidote to the Bayonet Attack in the King's Indian Defence, Mar del Plata branch.

I am writing this to do what I always do: open the board, read every line of the sales copy again, and check it against what actually happens on the board.

A system that lives by giving ground first

The King's Indian begins 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5. Black lets White take the whole centre, builds a fortress around the king, then strikes back at that very centre with ...e5 and ...f5. The Classical system continues 7.0-0 Nc6 8.d5 Ne7, locking the centre with the d5 pawn, and from there the two sides launch a two-wing race: White attacks the queenside, Black attacks the kingside. The name Mar del Plata comes from the 2026 tournament in Argentina, where the ...Nc6 and ...e5 structure in exactly this move order was brought into top-level practice; in ECO terms it sits around E97 to E99. David Bronstein was one of the players most closely tied to the system from the 1950s.

I still remember how I learned this system. In 2026, while working as a chess commentator for VTC, I spent two years in front of a board explaining to viewers why one side would let the opponent take the entire centre and still not flinch. Those two years taught me to read structures rather than moves: to look at empty squares, at pawn directions, at which wing would blow up first.

From the early 1990s, the Bayonet Attack — the move 9.b4 — became the favourite weapon of White players who wanted queenside space. Vladimir Kramnik was one of the leading players who brought it into top-level games, while Garry Kasparov repeatedly chose the Black side of this system in the same decade. White's idea is simple: push the pawn to b4, prepare b5 to clamp the queenside shut, then calmly transfer pieces to the kingside while Black hunts for counterplay.

How does Black answer? With 9...Nh5. The knight steps to the rim, attacks the f4 pawn, and clears the way for ...f5 and then ...f4. The game becomes a calculation tunnel: whoever is one tempo late loses the king. That is why 9...Nh5 became the main line, and also why it eats so much time for amateur players.

In the economics of modern chess, memory is a commodity. A professional carries thousands of moves from hundreds of lines in his head; an amateur with a few hundred lines is already full. Any product promising few concrete lines and little memorisation touches exactly that sore spot. That is fertile ground for recorded-lecture theory packages.

What 9...a5 does, and what it trades away

One thing must be said immediately: 9...a5 is a move that already exists in theory. It is not new. It sits among the secondary options and has coexisted with the main line 9...Nh5 for a long time. What is being sold here is a packaging, an emphasis, a way of telling a story — not an invention.

Mechanically, the pawn push to a5 puts a direct question mark on the b4 square White has just occupied. The a5 pawn attacks the b4 pawn and forces White to decide within a few tempi: push b5 to close the queenside and turn the structure into a wall, trade pawns to open a file, or keep the tension and accept that the queenside will be a place where Black has a voice. Each choice leads to a different kind of game, and that variety is precisely the product's selling point: Black is no longer dragged into a single tunnel.

In exchange, Black gives up the strongest thing about the main line. 9...Nh5 has a clear target and a clear countdown: ...f5, ...f4, open the kingside, drive White's king into defence. 9...a5 has no such clock. The game slows down. The two sides enter the kind of game in which, after move twenty, there are no more lines to memorise — only plans. For a player who likes to think independently, that is a reward. For a player used to leaning on memory, it is a punishment, because with nothing to remember you have to think, and thinking is far more tiring than remembering.

White also gains and loses. White keeps queenside space, keeps a closed centre, and on paper retains the initiative. But White loses a clear counter-target. When Black does not charge up the kingside, White's queenside counterplay becomes less urgent; both sides wait, both accumulate small gains, and the result usually comes from a weak square someone spots on move thirty-five.

Now to the sales copy. The product makes three claims: 9...a5 leads to sound positions with very few concrete lines; computers often do not know how to handle this kind of complex strategic position; and if Black knows what he is doing, White can quickly find himself in danger.

None of those three claims comes with data. No author name, no title, no sample games, no engine evaluation, no win-draw rates, no specific release date, no count of practical games. A verifiable claim would look quite different: for instance, in a given database, how many games feature 9...a5 above the 2600 level, what percentage White wins, what percentage is drawn, and how the engine rates the position after ten moves. That data is entirely available. It does not appear.

The most telling claim is the one about computers. The sentence about computers often not knowing how to handle complex strategic positions sounds very familiar — it belongs to the pre-AlphaZero generation. Before neural-network engines, analysts still believed machines were strong in tactical positions and weak in long-term strategic ones. That belief has been overturned. Today's engines assess closed King's Indian positions better than almost any human player, and they see twenty-move plans the human eye misses. A theory package advertising itself with that sentence is selling a premise that expired at least half a decade ago.

In 2026, in my data-analysis work, I had to throw away half an old dataset because I realised competitive conditions had changed so much that the old models no longer held. In chess, the equivalent marker is the arrival of neural-network engines. The earlier dataset was not absolutely wrong, but the way of reading it had expired. No tactic is ever old; only the way of reading the game expires. A sales line built on an old premise can still sell, but it will not win the buyer a single extra game.

There is one more detail that made me stop. The headline attached to the package describes a team round in Samarkand, suggesting that every strong team won despite close calls, while the entire body of the text discusses exactly one recorded opening lecture. Two unrelated texts were stitched into one package. To me that is a source-quality signal: if the product description itself is mismatched with its headline, then the credibility of everything else has to be checked from scratch.

An antidote only lives until the other side prepares

In chess, the notion of an antidote has a very short lifespan. It holds only until the other side opens a database, finds the rebuttal line, and brings an improvement to the board. The Bayonet branch has lived in theory for three decades and has been dissected move by move by leading players. A secondary move like 9...a5 cannot erase those three decades. It only opens one more door.

For a well-prepared White player, 9...a5 is not a nightmare. There are ways to handle it that keep the space and keep the initiative. The claim that White quickly gets into danger if Black knows what he is doing is an absolutist sentence — exactly the kind of sentence advertising copywriters use, and impossible to verify from the text itself. It does not say in which line White gets into danger, on which move, after which improvement. It only creates a feeling.

The most interesting part of the package is the sales pitch itself. It targets a self-image: the player who likes to think for himself, who dislikes memorisation, who does not want to become a memory machine. That positioning is clever, because it does not sell a line of theory, it sells an identity. The buyer is not buying 9...a5; the buyer is buying a picture of himself. And once emotion is touched, people demand less data.

But the price is real. An amateur spends an entire weekend on 9...a5, memorises the main branches, then goes to the club and meets an opponent who walks straight down the old road. The product is not wrong; it simply does not appear. The time spent does not come back.

In my first summer working as an analyst in the V-League, I learned something I now keep as a professional rule: a diagram is only beautiful when the opponent agrees to stand still. On a chessboard that is even truer. A move only has value when the opponent is forced to answer it. 9...a5 is a decent move, but it is only beautiful when White agrees to play Black's game.

There is also the buyer's side to consider. If a theory package names no author, cites no record by that author in the very system being taught, and offers no sample game with an engine evaluation, then the buyer is paying for a belief. The opening-theory industry makes its living selling answers to fashionable lines, and that is one of the most stable revenue streams in chess. Stability does not mean quality is always there. It only means there is always a buyer.

What to track, and the real value of the story

With a package like this, the real value is not in move nine. It is in the habits the buyer forms afterwards: whether he opens a database to double-check, whether he writes down the places where he got refuted, whether he dares to reread his own earlier judgments and strike them out.

I will track three signals over the coming months and set them down here so I can compare later.

What interests me most is the appearance of 9...a5 in classical games above the 2700 level. If the move shows up regularly, the package's premise is confirmed. If it stays in amateur games, it is only a beautiful line on paper. I am not asking it to win; I am only asking it to be played.

Alongside that is the frequency of 9.b4 in opening trees. The value of an antidote depends on whether the disease is still common. If the Bayonet Attack loses its fashionable status — entirely possible when opening trends rotate — then both the product and the debate around it fade on their own.

The remaining signal, and the slowest, is a rebuttal line from the White side. The cycle of theory is always this: one side offers an answer, the other answers the answer. Once 9...a5 is played often enough, someone will find an improvement. At that point the package's value depends on whether it gets updated — and this is where recorded products usually lag behind open databases.

Based on my experience following games, I think the King's Indian in the Mar del Plata branch survives at the top level, but as a surprise weapon rather than an everyday choice. In the engine era, people are wary of systems that concede space and invite the opponent to calculate. That is why this branch needs missionaries — and the package in question is, after all, a carefully packaged missionary lecture.

If you are an amateur considering spending money, there is a very cheap step to take first: open a free database, find 9...a5 in the opening tree, see how many games have been played, at what level, and with what results. If it appears often enough that you believe you will face it, buy. If not, keep your money and spend that weekend on endgame training — something that is certain to appear in every game you play.

On a chessboard, nobody can force you to believe a line of advertising. Moves are transparent, data is checkable, and players have the right to demand both before spending money. The value of 9...a5 is not in what a product calls it, but in whether it forces your opponent to sit back and think.