The fagIAno
Before every race that matters it says who can win and how likely it is. At the start the probabilities are frozen; after the finish comes the receipt, even when we got it wrong.
Against the experts · the 2026 test
Better than the experts in 30 races out of 41.
The test is run on races the model had not seen, against the newspaper predictions published before the start. The record below is something else: the forecasts frozen and published before the start, from Il Lombardia 2026 onwards, each with its receipt.
How it is measured: in the methodThe record
Every race with the probabilities frozen at the start and its receipt, even when it went wrong.
- 6
- races and stages with a receipt
- 5 out of 5
- races and stages with the experts’ prediction: the fagIAno gave the winner more than they did
- 3.10
- fagIAno’s mean log loss on the 5 races and stages with a newspaper prediction; the experts’ 5.51
The receipts
- No. 001 · frozen 10/09 22:53 Il Lombardia 2026 10 October · Bergamo → Como Seixas wins, 3rd on our list To the winner 11.3% Log loss 2.15 · Experts 3.15 SHA-256 fingerprint 290e1c088cf424d1b91d29828fedefbec3888864283ffcdc72f2aac9f94cb5c9 Image
- Frozen 10/12 15:14 Tour of Guangxi, stage 1 13 October · Chongzuo Magnier wins, 2nd on our list To the winner 21% Log loss 1.50 SHA-256 fingerprint 80e4ca3ab75675d3717a3660310024962dbae57150b308875d3761aae1a4c663
- Frozen 10/13 21:33 Tour of Guangxi, stage 2 14 October · Chongzuo → Jingxi Biesterbos wins, 15th on our list To the winner 0.7% Log loss 4.91 · Experts 6.05 SHA-256 fingerprint 49174f48bd347c8c26481bde5dadfdd0d347c69b79bbf6880bf25027c33603e1
- Frozen 10/14 21:33 Tour of Guangxi, stage 3 15 October · Jingxi → Bama Fretin wins, 8th on our list To the winner 1.8% Log loss 4.04 · Experts 6.13 SHA-256 fingerprint 1e8858032e5d1b87c04c56c02189e51af1fc7f215bac86514601a790fcfcae19
- Frozen 10/15 21:38 Tour of Guangxi, stage 4 16 October · Bama → Tian’e Magnier wins, 2nd on our list To the winner 23% Log loss 1.17 · Experts 6.11 SHA-256 fingerprint 398258e5b23228ec37689c133b0a8896ad4fc3b1a8c7eba484e9281e6e795f83
- Frozen 10/16 21:38 Tour of Guangxi, stage 5 17 October · Jinchengjiang → Yizhou Fretin wins, 5th on our list To the winner 3.9% Log loss 3.24 · Experts 6.10 SHA-256 fingerprint 95d9e79835728401dfa7cdb546ffbdf03c8646cca58986d306d1a42a0fa1c674
- Frozen 10/17 22:30 Veneto Classic 18 October · Cittadella → Bassano del Grappa The receipt comes with the official result SHA-256 fingerprint b299a0213aec3fd9f5cea0dbea5296a5b1063bac538570f47de2a31fc43bee56
Every race, as a table
| Race | Winner | We had given them | Top ten | Log loss fagIAno · experts · UCI · form |
|---|---|---|---|---|
| Veneto Classic 18 October 2026 | Frozen on 17 October at 22:30: the receipt comes with the official result. | |||
| Tour of Guangxi, stage 5 17 October 2026 | Milan Fretin | 3.9% #5 on our list | — | 3.24 · 6.10 · — · — |
| Tour of Guangxi, stage 4 16 October 2026 | Paul Magnier | 23% #2 on our list → 31% over the 145 starters | — | 1.17 · 6.11 · — · — |
| Tour of Guangxi, stage 3 15 October 2026 | Milan Fretin | 1.8% #8 on our list | — | 4.04 · 6.13 · — · — |
| Tour of Guangxi, stage 2 14 October 2026 | Frits Biesterbos | 0.7% #15 on our list | — | 4.91 · 6.05 · — · — |
| Tour of Guangxi, stage 1 13 October 2026 | Paul Magnier | 21% #2 on our list → 22% over the 150 starters | — | 1.50 · — · — · — |
| Il Lombardia 10 October 2026 | Paul Seixas | 11.3% #3 on our list → 11.7% over the 174 starters | 6 of 10 | 2.15 · 3.15 · 2.85 · 3.53 |
Calibration
When the fagIAno gives 20%, that rider should win about one time in five. Here the probabilities we gave, in bands, against how often the riders in each band actually won, over the 264 races of the version benchmark (v0.3).
The fagIAno versions
The method, in short
in use: v0.3
-
The list, before the start
For every race that matters, who can win and how likely it is, from the official UCI results.
-
The freeze
At the start the probabilities are frozen with the time and a SHA-256 fingerprint: after that they are never touched.
-
The receipt
After the finish, the comparison with the official result and with the newspaper predictions published before the start.
-
Judged over dozens of races
Log loss: minus the logarithm of the probability given to the winner. The lower, the better the prediction. One race says little, because even a good forecast often misses the winner.
The test against the experts, in log loss
- The fagIAno
- 2.85
- The experts
- 3.28
difference −0.42, 95% between −0.68 and −0.17
Log loss is computed on the starters: what we had given to riders who then did not start is spread among those who did. The experts are the newspaper predictions published before the start, turned into probabilities with a fixed recipe; where there is no prediction, there is no comparison.
-
The editor. Plans every piece, writes it, checks each number against the official results and gives the verdict: a story goes out only if it passes. After the finish it also makes the fagIAno’s receipt.
mimir.bot · Mnemosyne -
The voice. Mimir lets the fagIAno talk: it is the chat on every page, the same one that answers the question below.
mimir.bot
Ask the fagIAno
Races, riders, rules: it answers with our data.
Sources: official UCI results; newspaper predictions published before the start.