9...a5 and the Theory War Around the Bayonet Attack: When a Move Becomes a Product

Opening point

Three in the morning in Shanghai. On the screen sits a data file sent over by a news aggregation team. The headline is explicit: round 2 of a team event in Samarkand, all the strong teams winning even after close calls. The body of the file talks about a theory DVD teaching 9...a5 against the Bayonet Attack in the King's Indian Defense.

Two unrelated texts sharing one file.

I read it three times. No tournament appears in the body. No team, no board, no result, no player name. Only one move and one promise.

This is the kind of error that keeps me awake more than a good game. Not because it is rare. Because it is the common denominator of the era: chess theory content is mass-produced, packaged as a product, and then drifts through aggregation systems while nobody rechecks the body. When packaging fails at the headline level, my first question is not "is this move good or bad" but "what else in this file is wrong too".

Context

The King's Indian Defense runs 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. Black deliberately concedes the centre from the first moves, then strikes back at that very centre with ...e5 and ...c6 later. It is the purest surviving hypermodern system at the highest level.

The Mar del Plata variation takes its name from the 2026 Candidates Tournament held in the Argentine city of the same name. It is the sharpest branch of the system. Both sides race down opposite wings: White pushes pawns on the queenside, Black pushes pawns on the kingside, and the race to deliver mate is usually decided by move count rather than by feel. ECO codes E97 to E99 live here.

In the 1990s, the move 9.b4 appeared as a defiant answer and became known as the Bayonet Attack, a bayonet-style pawn thrust. White opens the queenside before Black can finish the kingside wall. White gains space and forces Black to choose: enter forced sequences, or concede the initiative.

Black's main reply is 9...Nh5. The knight jumps to the rim, opens the position, and generates concrete, long, highly forcing lines. Because of that forcing character, 9...Nh5 became fertile ground for engine analysis: everything is verifiable, and computable.

9...a5 is a sideline. Black pushes a queenside pawn, avoids opening the position early, and accepts a slower strategic fight in which the two sides contest weak squares instead of racing.

Thirty-two years in the commentary booth taught me one thing: viewers do not remember moves, they remember situations. Buyers of theory are the same. They are not buying a sequence of variations; they are buying the feeling that they already understand the position.

Core analysis

The product in that file is marketed as an antidote to the Bayonet. The central argument is tidy: while 9...Nh5 opens the position, forces the lines and makes computer analysis important again, 9...a5 leads to sound positions with very few concrete lines.

The appeal sits in four words: very few concrete lines. What is being sold is relief, not victory.

And this is where I stop, take out a pencil, and strike three lines through.

Not a single game is cited. No ACPL, the average centipawn loss. No win rate, no draw rate. No match rate against the engine's top choice. No author name. No record of the author in the very system he teaches. All that exists is a qualitative claim: the position is balanced, and Black quickly becomes dangerous for White if he knows what he is doing.

For anyone who works with data, this is a large void. A theory DVD that wants to persuade me must supply at least one of three things. First, the frequency of 9...a5 in elite-level databases, with a timeframe attached. Second, empirical results broken down by rating band, so we know whether the branch lives off strong players or survives only at club level. Third, engine evaluations at moves fifteen to twenty-five of the main lines, where the true nature of the position shows itself once the forced moves are gone.

All three are absent.

That does not mean 9...a5 is wrong. It means the claim is unverifiable, and the buyer is paying for a belief rather than for a calculation.

I once trusted emotion, until a number knocked on my door at three in the morning.

In this case, the number does not knock. It is simply absent. And that absence is itself data.

A little more context is needed, because it determines the real value of the product. The King's Indian at the top level has narrowed considerably in the engine era. Systems such as the Grünfeld or the Nimzo-Indian, with lower theoretical risk, account for most elite games. The KID survives, but as a surprise weapon: something a player prepares for a decisive game, not something played all season.

Against that backdrop, the Bayonet Attack remains the most unpleasant option White can produce. It does not need to win on theory. It only needs to make Black remember.

Which is why a product that teaches how to avoid remembering has a real market. That market is small, stable and loyal: club players who want a weapon without spending dozens of hours copying a database. That is why the product exists. Not because 9...a5 is stronger than 9...Nh5.

I light a candle for data. But I always let the flame of emotion light the way to the question.

The counterintuitive angle

Now for the twist.

The marketing rests on an outdated premise: that computers cannot handle complex strategic positions well. That sentence was once true. It was true before neural-network engines appeared. After December 2026, when AlphaZero played its famous exhibition match against Stockfish, and after the wave of neural-network engines washed into almost every analysis program, that gap was effectively erased. Modern engines assess long-term KID strategic positions far better than the advertising line implies.

In other words, someone is selling an advantage that has expired.

But the real twist lies elsewhere, and it is subtler.

If 9...a5 genuinely offered few concrete lines, why has it not become the main reply? The plausible answer: because it does not really offer few lines. The Mar del Plata branch remains sharp by nature, and no move removes the need to calculate. What the product sells is not calm, but the feeling of calm. Those are different things, and the gap between them is where buyers lose time and money.

Then comes the most telling detail in the positioning: the product targets "players who like to think for themselves, not memorizers". This is identity-based selling. It does not tell you about the quality of the variations. It tells you about the person you want to be. That approach works commercially, but it answers none of the questions a real buyer needs: what is move eighteen, and what is the evaluation there?

A pressing move is not noise. It is a question that data is whispering.

And finally, the biggest problem of all is not about chess.

One headline speaks of round 2 of a team event in Samarkand. One body speaks of a theory DVD. Two unrelated contents share one package. For someone who works with data and markets, this is the most serious signal in the entire file. If the packaging fails at the headline, nothing guarantees that the theoretical verification inside is sound. A move can be good. A mismatched data file is always bad.

Forward-looking takeaway

So what should be tracked in the next cycle?

First, the appearance of 9...a5 in classical games above 2700 Elo. If it appears, the product's premise is confirmed or refuted by real data rather than advertising copy. Second, the frequency of the Bayonet Attack in opening trees; if White keeps playing 9.b4 regularly, the product's window of relevance stays open. Third, and most importantly, the name of the person behind the line.

There are players who get forgotten, but data never forgets them. In this file, nobody is named. And a move with nobody behind it is still just a move.

The question left for the reader: if this sideline really neutralises the Bayonet, why does nobody in the 2700+ group trust it enough to play it in a rated classical game?