FRITZ 20: ChessBase Sells a Training Revolution, the Chess Market Wants Evidence

In November 2026, in Bonn, I watched Vladimir Kramnik allow mate in one against Deep Fritz in a press room that went silent. The machine saw the move in milliseconds. The human did not. Nineteen years later, in Chengdu, I opened a ChessBase product release for FRITZ 20, which describes itself as a chess engine and, above all, a training revolution. The document contains no engine strength figures, no price, no release date, and no feature list, only adjectives. Twenty years ago that sentence sold software. Today it does not, and the reason lies in the market rather than the product. ChessBase is a historic German company tied to the digitisation of chess databases, and the Fritz engine, developed by Frans Morsch and Mathias Feist, drew 2-2 with Garry Kasparov in New York in 2026 and beat Vladimir Kramnik 4-2 in Bonn in 2026. But the market FRITZ 20 enters is unrecognisable: Stockfish, the open-source engine, leads most ratings at roughly 3,600 Elo; Leela Chess Zero proved in 2026 that neural networks can reach superhuman strength without a hand-built opening library; Lichess offers free Stockfish analysis in a browser tab; and Chess.com has turned game review into a subscription. The promotional text raises three questions it never answers. How strong is the engine compared with free alternatives? What training mechanism distinguishes it from an ordinary analysis tool? What price justifies paying when the free option already serves most players? It also contradicts itself on audience, claiming to serve both tournament-level players and absolute beginners, two groups whose needs pull design in opposite directions. And the source is the manufacturer, with no independent verification, no publication date, and no named author. I stopped believing in miracles on the board years ago; I believe in conversion rates. A genuine training revolution must be provable through four measurable indicators: average centipawn loss per move, high-severity blunder frequency per hundred moves, opening retention tested after thirty days, and error rate under time pressure. I have tracked these on my own games since 2026, and the data told me something counterintuitive: most of my lost points did not come from badly played games, but from spending too long on trivial moves in the first seventy minutes and leaving too little time for the decisive ones. Software that calls itself a personal trainer must detect that pattern and prescribe exercises for the exact gap. Stronger analysis is easy. Correct diagnosis is hard, and diagnosis is absent from the FRITZ 20 material. This matters because the strongest engine is not the best coach. A 3,600 Elo engine offers advice a 1,500 Elo player cannot execute, not because he fails to understand it, but because acting on it requires a different order of priorities. The distance between the best move and the best move a human can play consistently is a distance of pedagogy. Other sports have already walked this road. After swimming banned polyurethane suits in 2026, world records stopped falling in waves and competitive advantage moved from equipment to training methods. Athletics went through something similar when carbon-plate shoes were restricted, pushing teams back toward load management. Chess now sits at the same junction. The tools have hit a technical ceiling; the way humans learn has not. There is also a business dimension. ChessBase grew on boxed software, and that model is squeezed from both sides: free tools on one end, subscription platforms on the other. A boxed product must justify a large one-time price, or switch to subscriptions. The release does not say which model applies, and that omission matters, because pricing shapes design. A one-time purchase only needs to impress in the first thirty days; a subscription must create a reason to return every week. My contrarian reading is this: free engines did not devalue paid training software. They increased its value, by clearing away the part that can no longer be sold. When calculation is given away, the only remaining product is structure, discipline, and timely feedback. Companies with a genuine technical edge publish benchmarks. Companies without one publish adjectives. Selling software is the one business where people pay for hope and blame time for the result. I am not calling FRITZ 20 a bad product. I have not tested it and I do not judge what I have not measured. What I am saying is that this release says far more about the seller's market position than about the product's quality, and that readers should treat every descriptive claim as unverified until independent data appears. Any player can run a thirty-day test: pick thirty tactical positions from your own lost games, record accuracy and time on first attempt, then revisit them on days seven, fourteen, and thirty, and compare against thirty control positions seen only once. If the paid software helps you remember better than the control group, you have your own evidence. If not, you have saved money. The final test is not the software. It is the user. What I want to see in the next six months is not a new version but a few pages from ChessBase itself: how many users complete a full training programme, how much their average centipawn loss falls, how their error rate under time pressure improves after sixty days. If those numbers appear, the story changes completely and I will be the first to rewrite this piece. If they do not, buyers should ask one simple question: am I paying for a tool, or for the feeling that I am improving?