How accurate are GPHG prediction models?

On the 2026 nominations, the GPHG Index correctly forecast 43 of the 84 nominated watches (51%) and publishes every pick and every miss. Free Sprung reports 55 of 84 (65%) but keeps its predictions sealed in password-protected PDFs until after the 7 November ceremony. Both comfortably beat the roughly 37% a random guess would score.

I'm not the only one predicting the GPHG

Q: Is there a model that predicts GPHG winners?

More than one now. The GPHG Index (this site) and Free Sprung both forecast the Grand Prix d'Horlogerie de Geneve from its own 25-year record of entries, nominations and prizes. Neither can know the winner before the jury votes on 1 November 2026.

On 18 June 1858, Charles Darwin opened his post to find a short essay from a young naturalist named Alfred Russel Wallace, working alone in the Malay Archipelago, that set out — clearly, and in complete ignorance of Darwin's twenty years of hoarded notes — the theory of natural selection. Darwin was, by his own account, floored. The one idea he had believed uniquely his own had occurred, independently, to a stranger on the other side of the world.

I had a very much smaller version of that feeling last week. I have spent an embarrassing quantity of time turning the Grand Prix d'Horlogerie de Genève — the watch world's Oscars — into a database and training a model to predict its results, on the quiet assumption that nobody else was daft enough to bother. Then, while asking six AI chatbots who would win, one of them mentioned a name I didn't know: Free Sprung.

Free Sprung is a Substack run by someone who has done almost exactly what I have — scraped 25 years of GPHG records (5,698 watches to my ~5,800) and built a model to forecast which entrants get nominated, and which nominees win. Reading it was like turning a corner in a very small, very obscure museum and finding another visitor already there, taking notes. Reassuring and faintly deflating in equal measure. Whoever you are: welcome.

Two models, one very niche question

We are, it turns out, asking the machine the same thing — of the watches competing in a category, which six will the Academy nominate? On the 2026 edition, the one live test we can both point to, Free Sprung reports catching 55 of the 84 nominees, or 65%. My own model managed 43 of 84, or 51%. Both of us land miles clear of the ~37% a coin-tossing monkey would average — and we arrived at that 37% baseline independently, by the same method, which is the sort of quiet agreement that makes me trust the baseline rather more than either of our headline numbers.

So on the face of it, Free Sprung is the better forecaster: fourteen points and a dozen nominees ahead of me. On the face of it.

The part I can't check

The difference between us is that my predictions are published in full — every watch, every score — and when I marked them I printed the misses beside the hits. Free Sprung's 2026 predictions are sealed inside password-protected PDFs that nobody can open until the author releases the passwords after the November ceremony.

To be fair, that is honest, and rather clever. A dated, frozen, downloadable-but-locked file is a genuine way to prove your forecasts existed before the results did — a commitment device I rather wish I'd thought of. But it cuts both ways. Until those passwords appear, that 65% is a figure I'm asked to take on trust: I can't see which watches were chosen, can't check them against the actual nominations, and can't confirm the per-category fields were drawn the same way mine were before the hit rates are compared. The most eye-catching claim of all, five of six correct in Essential — a category my own model bombed, managing only one of six — is precisely the one I can't yet verify.

None of which is an accusation. It's the ordinary gap between "trust me" and "here, look." I've chosen to show my working and my failures; Free Sprung has chosen to lock the answers in a box and pass round the key later. Both are perfectly defensible. Only one of them can be checked today.

Wallace and Darwin, once they'd recovered from the shock of one another, had their work read jointly to the Linnean Society and stayed cordial for the rest of their lives. I'd happily settle for that. The passwords come out after 7 November; I'll open the box, run the figures, and report back — on whichever of us turns out to have read the watches better.

In brief

Is there a model that predicts GPHG winners? More than one now: the GPHG Index (this site) and Free Sprung both forecast the Grand Prix d'Horlogerie de Genève from its own 25-year record. Neither can know the winner before the jury votes on 1 November.

How accurate are they? On the 2026 nominations, the GPHG Index caught 43 of 84 (51%) and publishes every pick; Free Sprung reports 55 of 84 (65%) but keeps its picks sealed until after the 7 November ceremony. Both beat the ~37% a random guess would score.