9...a5 and the Preparation Invoice: Auditing a King's Indian Opening DVD

I opened my inbox at eleven at night, after closing the day's data tables. The blurb ran less than a page: a King's Indian opening DVD, an anti-Bayonet system, marketed as an antidote for Black. I read it three times and went back to the database. Not a single eval. No ACPL. No game count, no author name, no titled player standing behind it. The page spends four lines insisting that computers do not understand these positions, and not one line telling the reader how many games the move has been played, at what level, with what score.

People call that a shock; I call it unread data. In this case, the data was never entered at all.

The position in question sits in the Mar del Plata, the heaviest branch of the King's Indian. White plays 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, then 9.b4 — the move that gives the variation its name. White pushes the queenside, aims at c5, prepares Ba3 and Nd2. Black must choose: strike the centre with ...f5 and ...Nf4 in the classical mould, or accept cramped space and counter with ...c5 on his own schedule. Both demand an enormous theory load, and that is the crux: White has ground that load down at the top level for three decades, with players such as Kasparov treating the King's Indian as a manifesto against the opponent's space.

The blurb proposes 9...a5 instead of 9...Nh5. The move is real, it is in the database, it is simply rare. It hits b4 directly and forces White to decide at once: push b5 and the queenside locks while Black gets ...Nd7 and ...b6 against the centre; trade on a5 and Black owns the a-file while White loses a comfortable structure. The structural point is genuine.

What is advertised as superiority is something else entirely. The real value of 9...a5 is a reduction in the variance of preparation cost, not an increase in the objective merit of the move. That is a legitimate practical benefit, mislabelled. For a club player with three hours a week for books, moving from memorising forty games to understanding three plans is an enormous saving. Saving cost, unfortunately, does not make the move stronger.

This is where the blurb hits its weakest link: the claim that computers do not understand these positions. That is a legacy of a bygone era. Since neural evaluation (NNUE) became standard — Stockfish 12 from 2026 onward, and the Leela family after it — closed strategic positions are no longer an engine blind spot at the level the text implies. The argument that survives the filter is narrower and different: 9...a5 positions contain fewer forcing lines to memorise, so the human-versus-engine gap is a memory gap, not an understanding gap. The engine can check the position; what it cannot do is make the player remember everything it checked. White's team can still run an engine for fifteen minutes and know the lay of the land. What they lack is the time to play it correctly inside those fifteen minutes.

The second claim deserves an entry in the ledger: that the chances are equal. No eval, no game reference, no database citation. In practice these Mar del Plata structures give White durable space and a queenside majority plan while Black plays for the initiative — a high-variance balance, not a symmetric one. At engine depth the number may be roughly level. But an unsupported assertion must be flagged as unverified, even when it happens to be true.

One small detail matters more than every technical claim: the move sequence in the blurb is broken. It stops at 7.0-0 and jumps straight to 9.b4, silently omitting 7...Nc6 and 8.d5 Ne7. Without those two moves the Bayonet does not exist. The fault lies in the documentation, not on the board — and it says something about the author: the text was written to persuade, not to be carried to a board and executed.

Then comes the problem the blurb never touches, though it is the product's central technical risk. The asymmetric payoff of a rare variation disappears the moment the opponent's preparation catches up. The half-life of surprise is inversely proportional to the frequency of the move, and that frequency never stands still. Every publicly played 9...a5 game decays a slice of the advantage that was purchased. The better the product sells, the faster the antidote expires — a commercial paradox no producer wants printed on the cover.

I tried to price the thing by rating band, the one part of the blurb whose logic I can reconstruct. Below 2026, opponents rarely build a proper Bayonet, so strategic guidance beats memorised theory: medium-to-high value. Between 2026 and 2400 is the peak, where opponents genuinely prepare and a rare move pays the largest surprise dividend. Above 2400, seconds and analysis teams already have the move in their files, the rarity premium collapses, and 9...Nh5 reasserts itself. Three bands, three values, and the blurb names none of them.

Another gap cannot be waved away in this product category: no author name, no title, no sample games from a named player. The genre norm is author credentials plus real games. That absence is not neutral: it leaves the buyer unable to judge the single most important variable — whether the product fits his own style.

The paradox sits here. The player who most needs this product is the one losing to better-prepared opponents; the one losing to a better player will not be saved by an opening DVD. Two different diseases, one prescription. Get the diagnosis wrong and the buyer ends up with another system, a dozen new plans, and the same loss in the same place.

One more point few notice: low-theory variations are expensive on the clock. In rapid and tiebreaks, a move that demands judgement every turn costs far more than a forcing line already memorised. The buyer is freed from memory and taxed on thinking time — a tax that only shows up when the clock drops under five minutes.

Numbers are asceticism: you have to give up comfort before you can see clearly. The transfer market is the only place where people pay for numbers nobody has verified — and the opening-book market is looser still, because there they pay for numbers that do not exist.

What I want to see over the next twelve months is data, not praise or blame: the frequency of 9...a5 in public databases, its actual score by Elo band, and how White answers with b5, Ba3 and Nd2 — three directions the blurb ignores entirely. If the frequency rises while Black's score does not, what was sold is a brief silence in opening history, and the antidote remains unfound. If the score improves between 2026 and 2400 and flattens above it, we will have the cleanest proof of an old rule: openings do not win games, preparation time does — and there is always someone willing to pay more for time.