Between 9...Nh5 and 9...a5: The Theory War Around the Bayonet Attack

A Thursday evening at a chess club on Nguyen Thi Minh Khai Street, District 3, Saigon. I sat behind a sixteen-year-old student of mine, watching him take the black pieces. The game followed the familiar road: 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 O-O 6.Be2 e5 7.O-O Nc6 8.d5 Ne7. The King's Indian Defence, Mar del Plata variation. On move nine, White pushed 9.b4 — the Bayonet Attack. My student thought for nine minutes, chose 9...Nh5, then lost two queenside pawns inside twelve moves and went down before his opponent's clock reached thirty. He turned and asked whether there was another way. Three days later, a promotional mailer for an opening-theory DVD landed in my inbox. The product answered exactly that question: an antidote to the Bayonet Attack, built on the ninth move 9...a5. The description was written with great confidence. It was also very short on data. To understand why a ninth move deserves an article, it has to be placed on its own branch of the tree. The King's Indian is the classic hypermodern defence: Black concedes the centre to White, builds a fortress on g7, then counterattacks the very centre he just gave away. Within that family, the Mar del Plata variation — Black plays his knight to c6 and his pawn to e5, White pushes d5, then Black pushes f5 — is the sharpest branch of all. The name ties to the tournament held in Mar del Plata, Argentina in 2026, where the leading grandmasters of the day put it under the knife. By classification, the system sits inside the ECO codes E97 to E99, and that is the only lookup-verifiable fact attached to this whole story. The Mar del Plata is a game of two wings advancing at once. White pushes b4, c5 and b5 to strangle the queenside. Black pushes f5, f4 and g5 to crush the kingside. Whoever arrives first wins; whoever is one tempo late gets buried. It survives at every level, even though at elite level its share of classical games has fallen sharply since the 1980s and 1990s. That decline has nothing to do with the line getting weaker; it comes from preparation costs rising while the expected return stayed flat. When the Bayonet Attack — 9.b4 — resurfaced as a carefully drilled choice among the elite, Black had to choose. Either accept theory that has been dissected to the bone with 9...Nh5, or look for a detour. The DVD I read chose the detour: 9...a5. According to the promotional text, the sales case comes down to a few points. 9...Nh5 opens the position, forces the lines, and pushes computer analysis back into the foreground. 9...a5 leads to sound positions with very few concrete lines. The product targets players who like to think for themselves rather than memorise, and it promises that if Black knows what he is doing, the position quickly becomes dangerous for White. It sounds fine. What stands out is the absence of a single game, a single number, or a single name that would allow verification. One more detail has to be raised at once. The original item I read carried a headline about a team round report in Samarkand — R2 (open): Despite close calls, all top teams grab wins — while its entire body was a promotional write-up for an opening DVD. Two different texts had been spliced together, and the reader has no way to tell which part is genuine. For a writer, that is the first signal to remember: everything quoted from it has to be re-verified at the source. Mechanically, 9...a5 does one thing: it asks White's b4-pawn what it intends to live on. In the Bayonet, the b4-pawn is the spine of the whole queenside plan — it opens the way for c5, for b5, for an eventual rook lift along the b-file. Black asks the spine directly instead of swinging around to the kingside. Measured against 9...Nh5, this is a quantifiable trade. 9...Nh5 sends the knight to the kingside rim, heading for f4, clearing the way for ...f5 and turning the game into a two-wing race. Both sides know what they must play for the next twenty moves, and precisely because of that, the gap between the heavy memoriser and the light one widens. 9...a5 is different. Black does not commit a knight, does not commit a tempo; he commits a structure: keep the queenside closed, wait for White to define his own position, then play chess. The price of that choice is that White loses nothing at all. White keeps the queenside space, keeps the closed centre with the pawn on d5, and keeps the right to choose when to open the wing. Black trades some forcing power for some flexibility. In other words, Black does not reclaim space; Black refuses to let White convert space into an immediate advantage, and those are two entirely different jobs. Iran in 2026 did not defend with numbers; they re-established space metre by metre. Black with 9...a5 is doing the same. Fewer pieces on the queenside rather than more, just a re-division of the squares, forcing White to spend one tempo answering the question on b4 before he can think about c5. A single tempo in a closed position is sometimes as expensive as a pawn in an open one. The most telling claim in the promotional text is the one suggesting computers often do not know how to handle these strategically complex positions. That used to be true. Before neural-network engines arrived, analysis tools were strong in positions with many forcing lines and combinations, and weaker in positions whose value lies in long-term structure — exactly the kind the King's Indian produces. But that threshold passed long ago. Today's engines handle long-horizon strategic positions far better than the previous generation, and the claim reads like marketing language from a bygone era. This is where I remind myself of something. In 2026 I threw away half of my old dataset, because football after the lockdowns was a different sport. Chess works the same way: datasets from before and after the neural-network era cannot be used interchangeably. A sales line written when computers still failed to grasp strategic positions now carries value only as an emotional invitation. The product sits in a very old and very stable market: selling answers to popular lines. Opening DVDs, online courses, paid databases — all of them live off a very human fear, the fear of being caught out on move fifteen. Every time a line becomes fashionable, the market produces an antidote, and every antidote produces an antidote to itself. That loop has no end and no beginning. What is notable is that products of this kind compete directly with free alternatives. Open databases, open-source engines, public studies on online platforms — all of it is already there. The real value of a DVD is not in the moves but in the curation and the narration. Buyers pay for a route, not for a variation. In Vietnam, the user base for this kind of product is mainly club players and students inside training academies. Top players such as Le Quang Liem or Nguyen Ngoc Truong Son have their own support teams and their own databases; they are not the customers of a retail DVD. What is being sold is not a weapon for the elite, but a tool so that an amateur stops being bullied on move nine. And that is precisely where the promise of very few concrete lines becomes the most attractive promise of all. For an amateur, the most expensive thing is not a good move but time. A variation demanding twenty memorised moves has no place in a working person's training schedule. A variation demanding structural understanding does. In return, 9...a5 demands something far harder to teach: the ability to play chess in a closed position where there is no combination to memorise. White will have space, will have more routes forward, and will have more ways to ruin his own position. Black must wait patiently for the right moment to open the kingside, and must accept that for the first fifteen moves he has nothing to gain. That is a skill no DVD can sell; it can only be bought with lost games. A heat map can lie, but five consecutive failed presses cannot. On the board, the equivalent test is five occasions on which Black fails to generate a single real threat on the kingside while White keeps advancing on the queenside. If that repeats often enough, the so-called sound position is just a slower way to lose. The lesson from that V-League summer still holds when placed on a chessboard: a scheme is only beautiful while the opponent agrees to stand still. The Bayonet tree is only beautiful while White agrees to play the exact tree Black prepared for. Once White changes tempo, changes direction, changes even the way he asks the question, all the preparation has to be redone from scratch. That is why every antidote has a short shelf life, and buyers should know whether they are purchasing a season or a decade. A note on data limits. The text I read carried no clear publication date, no author name, no title, no sample games, no engine evaluation, no win-draw rates from any database. All four of its sales claims are qualitative. That does not mean they are wrong. It only means they are unproven, and the buyer pays the price for the unproven part. A decent opening product normally ships with three immediately checkable things: the author's identity and record in the very system being sold, a handful of sample games with engine annotations, and an explicit statement that this is a known branch rather than a new invention. All three are missing here. The most obvious blind spot is that the product positions itself as an antidote. In opening theory, no antidote lasts long. A ninth move once regarded as a sideline becomes a main line the moment White publishes one good analysis, and the moment it becomes a main line it needs a new antidote. A secondary branch becomes stronger because it is under-prepared, not because it is better. And being under-prepared is a time-limited advantage, usually measured in months. At another level, the claim that computers do not understand carries a consequence the copywriter probably never considered. It teaches the buyer an anti-scientific habit: trusting human intuition before checking with a tool. In chess, that habit has been paid for dearly in games where a player believed in his beautiful position while the engine had already spotted a winning shot on move thirty-seven. Intuition is a starting point, not an ending point. The third blind spot sits entirely off the board, and it matters more than the other two. The original item was published with the wrong headline. Readers have no way of knowing which part is real and which part is an editorial technical error. For an analytical piece, that is an information-integrity problem: without a verifiable source, every conclusion drawn from it is provisional. In this profession, I have learned that a self-contradicting source is never the reader's fault. And I keep coming back to an old line: no tactic is ever outdated; only the way we read the game expires. 9...a5 is not outdated. It has been in the books for a long time, and it can still be a good choice for a Black player who knows what he is playing. What has expired is the packaging of it as a solution that needs no verification. The question is not whether 9...a5 is stronger than 9...Nh5. The question is how many games at the 2700+ level will adopt it over the next twelve months, and once it appears, how long until White has an answer. If it never appears, then this product is just a book chapter sold to club players, and there is nothing wrong with that — as long as buyers know what they are buying. While we wait, a sixteen-year-old student in Saigon will still sit at the board and ask himself: am I memorising a variation, or understanding a position?