The 9...a5 move against the Bayonet Attack: reading an opening-theory DVD in the engine era

On a Tuesday evening, at a chess cafe on Nguyen Thi Minh Khai street, an eleventh-grade student reset the pieces on the board and asked me a short question: playing Black in the King's Indian Defence, what should he do against the ninth move 9.b4. I answered out of professional reflex: 9...Nh5, that is the main road, the material is everywhere, just learn it. He nodded and wrote it in his notebook, then asked: "And if I don't want to memorise?"

Before I could answer, my phone buzzed. An analysis file arrived. I opened it, and the headline read "Samarkand R2 (open): Despite close calls, all top teams grab wins" - a round-two report from a team event, the gist being that the strong teams all won narrowly. Yet the entire body below, from the first line to the last, discussed an opening-theory DVD: the move 9...a5 against the Bayonet Attack.

The two texts have nothing to do with each other. One tells a story about a team in Uzbekistan, the other sells a chess disc. Across seventeen years of note-taking, I have learned an uncomfortable lesson: a mismatch between headline and body is the worst kind of error in a data file, because it does not knock you down at once, it only makes you wrong slowly. So that night I sat back down, and the lines below are my notes on what sits between those two pieces.

Context: why 9.b4 has become a headache

The King's Indian Defence, Mar del Plata variation, is one of the thickest remaining bodies of theory in classical chess. Black plays ...Nf6, builds the pawn wall g6, places the bishop on g7, then hands the centre back with ...d6 and ...e5. White takes space with d4, c4, e4, pushes the pawn to d5 and locks the centre. Black does not object. Black concedes both the centre and the space in exchange for two counter-attacking wings: White attacks on the queenside, Black attacks on the kingside. This is the kind of opening books call hypermodern - concede first, occupy later.

The main trunk of this variation runs exactly nine moves: 1.d4 Nf6 2.c4 g6 3.Nc3 Bg7 4.e4 d6 5.Nf3 0-0 6.Be2 e5 7.0-0 Nc6 8.d5 Ne7. From here White has two major directions: bring the knight to e1 and push f3-f4, or play 9.b4 - the move analysts call the Bayonet Attack. The b-pawn advances, threatening b4-b5 to drive the c6 knight away and open a path for the bishop to a3 or c1. White grabs queenside space first, then turns to the kingside.

In the ECO catalogue, this branch sits around E97 to E99, depending on the compiler's classification. That range is my own classification, not a note found in the original document.

The orthodox reply to 9.b4 is 9...Nh5. The knight jumps to the rim, threatens ...Nf4, clears the way for ...f5, and most importantly forces the position open. After 9...Nh5 both sides are pushed into concrete move-by-move calculation, because every mistake costs material. The lines here are long, and misremembering one move loses a wing.

9...a5 is a different road. The a-pawn advances, attacks the b4-pawn directly, and contests the very space White has just gained. Black does not hurry. Black steers the game into a more closed shape, where both sides must build long-term plans square by square rather than recite a forced sequence.

That is exactly what the DVD is selling.

The move 9...a5: an old branch repackaged as a new promise

The first thing to state clearly: 9...a5 is no invention. It is a line that has long existed in the theory archives, known to professional players as a fallback option, never as the main road. In the analysis file I read, not a single sample game is cited, there is no author name, and there is no engine evaluation. The product copy offers exactly one argument: this move leads to a balanced position, with few forced lines, so the player does not need to memorise much.

It sounds reasonable. But two separate things must be pulled apart. The first is objective theoretical value: does 9...a5 really balance the position? The second is practical value: does it help a player at some level survive the game? Product copy tends to merge these two questions into one, and that is where the buyer gets steered.

Technically, 9...a5 has a clear logic. White has just spent a move pushing the pawn to b4; Black replies by attacking that very pawn. If White pushes on with b4-b5, the queenside shuts down and Black gains time to complete the kingside plan with ...f5 and ...g5. If White holds, Black can play ...axb4 and then ...b5, trading pawns and opening the a-file for the rook. In both cases the character of the game shifts from "who remembers more" to "who understands the structure better".

That is an attractive promise, and I understand why it sells. At club level, most ordinary players lose not because they calculate poorly, but because they get dragged into lines their opponent has prepared more thoroughly. A move promising "less to remember" sounds like an escape route.

But here I have to be blunt: the promise of "few forced lines" is a description of the position's character, not a guarantee of the result. The copy never claims 9...a5 is theoretically stronger than 9...Nh5. It only says both sides have equal chances. That is a positional statement, not a claim of superiority.

What the DVD is really selling

Reading the introduction carefully, I realised the product sells three things, and all three sit off the board.

The first is a shift in the character of the game. From both sides charging into lines that must be memorised, to both sides building. For a club player this is a genuine benefit, not a gimmick.

The second is a stance against memorisation. The copy states plainly that the product is for "players who want to think for themselves instead of learning like a machine". This is an identity-based sales angle: it does not promise you will win more, it promises you belong to a different group of players. I have worked long enough to recognise this formula. It works, and it also hides a simple fact: in chess, understanding structures and remembering lines are not mutually exclusive; the strongest players do both.

The third is a counterweight to computers. The copy says that after 9...Nh5 the position opens, the lines become forced, and computer analysis matters again; whereas after 9...a5 you get sound positions with very few concrete lines, to the point that computers often do not know how to handle them.

That last sentence is the one worth discussing.

The blind spot: a computer story that is ten years old

The argument that "computers do not understand these positions" belongs to a specific era. Before neural networks appeared, common analysis tools ran on search and crude evaluation; they were very strong in forced lines and relatively weak in positions demanding long-term plans like the Mar del Plata. In that period the sentence sounded convincing.

But chess passed that milestone long ago. Since neural-network programs took over, and especially since NNUE-style engines became widespread, the machine's ability to handle long-term positional play has changed completely. Closed positions with few forced lines, where humans have to "feel" the plan, are precisely where modern engines look frighteningly superior, because they can evaluate space that the human eye cannot see in full.

Esports taught me something football hides well: a reaction within 200 milliseconds can break an entire system. In chess that milestone is longer, but the principle is identical: once the evaluation tool changes, the whole old theoretical ranking gets scrambled. If an opening course still uses "computers do not understand this position" as its main selling point, that selling point expired at least one engine generation ago.

In 2026 I threw away half of my old dataset, because football after the lockdown was a different sport. I mention that not to compare the two games, but to describe a habit: when the measurement foundation changes, every conclusion built on the old foundation must be re-read. Chess went through exactly that foundation shift, it just happened more quietly.

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

Not a single metric to verify

The most uncomfortable part of this file is the emptiness.

A serious opening-theory product normally comes with at least three kinds of data: the win-draw-loss rate of the line in a database, the degree of agreement with engine choices, and a few sample games at a high level. The file I read has none of the three.

No author name. No title. No competitive results for the presenter in this very line. No engine evaluation for any move. No evidence about time controls, and no release date. For a buyer this means there is no way to distinguish a carefully compiled product from a rehash of material that is already free in open databases.

I still keep the habit of taking notes on site whenever I watch domestic tournaments. One pattern repeats year after year: Vietnamese players learn openings very quickly, but almost nobody checks the source. A video, a post, a word of mouth at the club is enough for a move to become "standard".

This is what I always write in my "Data limits" section, and here it matters even more: missing data does not mean the content is wrong, but it does mean you are buying a belief, not a verified conclusion.

The opening market: selling moves in the age of free data

There is an economic question behind the technical story. Why would anyone still pay for an opening course when every line of the Mar del Plata can be looked up for free, with engine evaluations, in a few clicks?

The answer is not in the data. It is in the structure. A database answers the question "has this move been played". Books, videos and courses answer the question "what does this move mean". Beginners are not short of information; they are short of priorities. A good compiler helps them know where to look first, what to skip, and where to stop and think.

That is the real value of this product category, and it is far more modest than the copy promises. It is useful for club players and international masters who need a map. It is nearly meaningless for elite players, who have their own analysis teams and update theory weekly.

Understanding where a product actually stands helps the buyer place expectations correctly. It is also why I do not read an opening DVD the way I read a transfer story. The transfer market is chess, not a card game - but many sporting directors like to flip cards. The opening-theory market works the same way: the seller knows the true value, the buyer usually sees only the display.

The trap called "the counterweight"

That V-League summer I learned one thing: a formation is only beautiful when the opponent agrees to stand still. In chess, that translates as follows: an opening idea is only beautiful when the opponent has not prepared for it.

The Bayonet branch is played widely at every level. That means many people have spent time finding ways to counter precisely the moves this DVD sells. An idea advertised as a "counterweight" today can become a target dissected tomorrow, simply because opening theory runs in cycles: someone proposes, someone refutes, and the next round begins.

That is the biggest blind spot of every product of this kind, and the one the copy never mentions: the value of a move lies not in how good it is, but in how many people are trying to counter it.

There is a small but telling detail here. The copy closes with an absolute sentence: if Black knows what to do, the position can quickly become dangerous for White. A sentence like that cannot be verified, because it depends entirely on how well White has prepared. It shapes the buyer's feeling without offering any variation to check against. To a data analyst, this is the classic signature of advertising copy, not analysis.

Headline and body mismatch: a worrying signal

Back to that Tuesday night.

The headline is about a team event in Samarkand, round two, where the strong teams all won. The body is about a DVD. These two cannot belong to the same document. For a data analyst this is the most serious red flag in the whole file, more serious than missing figures.

The reasoning is practical. If a data file has a headline spliced onto the wrong body, then nothing guarantees that the facts inside are intact. In this file the body describes a product with no author, no release date and no sample games. The head describes an event that never appears in the body. Both halves are unverifiable, and they do not belong together.

One possibility is worth considering: the Samarkand event may be real, and the DVD content may be real, but they come from two different sources merged during a data process. If so, the correct approach is to separate them and trace each half to its origin, rather than trying to reason about how they connect.

For an analyst, the right conclusion here is not "this product is good or bad". The right conclusion is: to evaluate the content, trace the primary source first.

Takeaway

I still check everything twice, including the things that sound pleasant.

Three things are worth watching in the coming months. First, whether 9...a5 shows up more often at the elite level - if it does, its promise has been confirmed by practice; if not, it stays exactly where it was, a fallback option. Second, the frequency of the Bayonet Attack 9.b4, because that measures the lifespan of any product built against it. Third, and most important to me: whether the chess community starts demanding verifiable data before paying for an opening course, or keeps buying on feeling.

The eleventh-grade student asked me a question I had not finished answering: if not memorising, then what. The honest answer is that no move is free of memory. There are only moves that make you remember less and, in exchange, understand more. The problem is that not everyone selling you that move makes the price clear.