Project F1
I’m running a timed AI experiment in the F1 prediction markets. Small stakes. Real ledger. Updated after every race.
Starting stake / per-race new capital: $5.00 · Contributions: $30.00 · Current value: ~$17.91 · Races completed: 6 (Dutch, Italy, Spain, Azerbaijan, Malaysia, Singapore) · Last-race net Δ: -$0.17 (Singapore) · Last updated: 13 October 2026 (Singapore settled)
3-race allocation test: complete (3 of 3)
Singapore (Marina Bay) · Singapore GP, 11 Oct 2026 · result
Verstappen won from pole, 7.335 s ahead of Leclerc, with Antonelli third.
Top 10: 1 Max Verstappen (Red Bull Racing) · 2 Charles Leclerc (Ferrari) · 3 Kimi Antonelli (Mercedes) · 4 Lando Norris (McLaren) · 5 Oscar Piastri (McLaren) · 6 Lewis Hamilton (Ferrari) · 7 Liam Lawson (Racing Bulls) · 8 Nico Hülkenberg (Audi) · 9 George Russell (Mercedes) · 10 Isack Hadjar (Red Bull Racing)
Official classification: https://www.formula1.com/en/results/2026/races/1296/singapore/race-result
Model check
(v1.3.0, 500,000 simulations, frozen before Kalshi):
Verstappen win 23.8478% (ranked #1, won) · Leclerc 16.8062% · Antonelli 15.5380% · Hamilton 14.9652%
Podium: Verstappen 49.8314% · Leclerc 44.6038% · Antonelli 41.0878% · Hamilton 43.4794%
Top 10: Verstappen 80.9778% · Antonelli 83.5956% · Hamilton 83.7238% · Leclerc 81.7876%
Scores: winner log loss 1.4335 · Podium Brier 0.06078 · Top-10 Brier 0.09188 (22 drivers). Picking the winner is not proof of calibration.
Positions (barbell, A-1.4.1)
| Position | Cost | Outcome | Payout | Net |
|---|---|---|---|---|
| Antonelli Top 10 YES | $2.50 | WIN (P3) | $2.67 | +$0.17 |
| Verstappen Top 10 YES | $2.50 | WIN (P1) | $2.65 | +$0.15 |
| Leclerc Winner YES | $0.25 | LOSS (P2) | $0.00 | -$0.25 |
| Hamilton Winner YES | $0.24 | LOSS (P6) | $0.00 | -$0.24 |
| Total | $5.49 | 2 wins / 2 losses | $5.32 | -$0.17 |
Fees ($0.01 per trade) are included in cost. Costs are from the holdings screen, not exact broker cash.
Sepang (Malaysia GP) · test race 2 of 3 · result
Results relevant to my positions: Max Verstappen P1 · Lewis Hamilton P3 · Charles Leclerc P4 · Isack Hadjar P5
Model check
(500,000 simulations, price-blind, from the pre-race Quant file):
Win: Verstappen 21.0% · Hamilton 15.5% · Antonelli 15.0% · Leclerc 13.0% · Russell 9.9% · Norris 8.3% · Hadjar 8.2% · Piastri 5.3%
Podium: Verstappen 47.0% · Hamilton 41.9% · Antonelli 40.3% · Leclerc 37.9% · Russell 32.8% · Hadjar 31.0% · Norris 26.2% · Piastri 20.6%
Top 10: Hamilton 83.8% · Antonelli 83.5% · Verstappen 81.9% · Russell 80.3% · Leclerc 80.2% · Hadjar 79.1% · Norris 74.2% · Piastri 73.3%
Model check vs result: the model's top win pick, Verstappen (21.0%), won.
Positions (barbell, A-1.4.x)
| Position | Cost | Outcome | Payout | Net |
|---|---|---|---|---|
| Hamilton Top 10 YES | ~$2.21 | WIN (P3) | $2.52 | ~+$0.31 |
| Verstappen Top 10 YES | ~$2.30 | WIN (P1) | $2.52 | ~+$0.22 |
| Leclerc Winner YES | ~$0.25 | LOSS (P4) | $0.00 | ~-$0.25 |
| Hadjar Winner YES | ~$0.25 | LOSS (P5) | $0.00 | ~-$0.25 |
| Total | ~$5.01 | 2 wins / 2 losses | $5.04 | ~+$0.03 |
The actual result is ~+$0.03, and that is what the ledger uses. (A normalized strategy-scoring figure exists in the Quant file, but it isn't a realized result and isn't shown here.)
Value after Sepang: ~$13.08 ($8.0488 + $5.0000 contribution + ~$0.03).
Race 1 of my 3-race test: done (4th race on the ledger)
Baku — Azerbaijan GP · result
Russell won and Leclerc finished in the Top 10, so both of my NO positions lost. My model had Russell as its #1 pick to win at 28.92%, while the market had him around 64%. I bet against the market's certainty, and this time the market was right.
| Position | Qty | Entry price | Fee | Total cost | Payout | Net P&L |
|---|---|---|---|---|---|---|
| Russell Winner NO | 23.32 | 37c | $0.38 | $9.0084 | $0.00 | -$9.0084 |
| Leclerc Top-10 NO | 4 | 9c | $0.02 | $0.3800 | $0.00 | -$0.3800 |
| Norris Top-10 NO | — | — | — | $0 | — | PASS (no position) |
| Total | $0.40 | $9.3884 | $0.00 | -$9.3884 |
Baku net: -$9.39 (53.8% of the $17.44 available; the uncommitted $8.05 was kept).
One race is not a verdict. I review the policy after all three test races (Baku, the next race, Singapore), not because of one result.
Amounts are calculated from the displayed receipt prices, not exact broker cash. The result rests on my own confirmation of the race outcome; final Kalshi settlement receipts and the S&P benchmark are still pending.
Baku: forecast locked before market check
Baku — Azerbaijan GP · pre-race (settled above)
No market price was used to generate or alter the probabilities.
Available ledger capital: $17.44 ($12.44 entering + $5.00 Baku contribution)Deployed: ~$9.39 (53.8%)Uncommitted: ~$8.05Race 1 of 3 in my A-1.3 allocation test
The model
I ran 500,000 simulations after qualifying. The model made Russell the favorite at about 29%.
Locked probabilities — every driver (%)
| Driver | Grid | Win | Podium | Top 10 | Finish |
|---|---|---|---|---|---|
| George Russell | 1 | 28.92 | 56.45 | 86.75 | 88.19 |
| Charles Leclerc | 2 | 16.83 | 44.39 | 84.06 | 85.81 |
| Oscar Piastri | 3 | 10.62 | 34.42 | 81.66 | 84.38 |
| Isack Hadjar | 4 | 7.22 | 26.56 | 81.24 | 85.91 |
| Lando Norris | 5 | 9.41 | 32.68 | 81.56 | 84.27 |
| Lewis Hamilton | 6 | 8.38 | 33.16 | 85.48 | 88.37 |
| Pierre Gasly | 7 | 0.59 | 3.73 | 62.78 | 90.10 |
| Max Verstappen | 8 | 8.97 | 29.11 | 80.68 | 84.83 |
| Carlos Sainz | 9 | 0.02 | 0.22 | 22.30 | 83.56 |
| Franco Colapinto | 10 | 0.23 | 1.70 | 47.29 | 91.59 |
| Oliver Bearman | 11 | 0.07 | 0.56 | 28.97 | 85.97 |
| Liam Lawson | 12 | 0.26 | 1.67 | 43.74 | 89.52 |
| Alexander Albon | 13 | 0.01 | 0.15 | 17.60 | 82.08 |
| Esteban Ocon | 14 | 0.04 | 0.34 | 23.35 | 90.42 |
| Arvid Lindblad | 15 | 0.29 | 1.76 | 42.67 | 89.48 |
| Kimi Antonelli | 16 | 8.00 | 32.07 | 83.36 | 88.93 |
| Gabriel Bortoleto | 17 | 0.07 | 0.51 | 22.83 | 87.69 |
| Nico Hulkenberg | 18 | 0.07 | 0.51 | 20.96 | 83.33 |
| Sergio Perez | 19 | <0.01 | <0.01 | 0.64 | 78.90 |
| Valtteri Bottas | 20 | <0.01 | <0.01 | 0.22 | 78.94 |
| Fernando Alonso | 21 | <0.01 | 0.01 | 1.56 | 76.48 |
| Lance Stroll | 22 | <0.01 | <0.01 | 0.30 | 70.71 |
Model estimates locked before the market check. They are estimates, not certainties.
Market check (after the lock, ~15:10 UTC)
After I locked the forecast, the public Kalshi pages showed:
- Russell Winner YES: ~64c
- Russell Winner NO: 37c
- Leclerc Top-10 NO: 9c
- Norris Top-10 NO: 10c, then 11c when I went to buy
Caveat: these are public page prices, not verified order-book snapshots.
The market showed Russell around 64%; my model had him at 28.92%. That disagreement led the model toward Russell NO. The model still ranks Russell as the most likely winner — the position is against the market's degree of certainty, not against Russell being the favorite.
Positions I bought.
| Position | Qty | Price | Fee | Cash at risk | Status |
|---|---|---|---|---|---|
| Russell Winner NO | 23.32 | 37c | $0.38 | $9.01 | Bought |
| Leclerc Top-10 NO | 4 | 9c | $0.02 | $0.38 | Bought |
| Norris Top-10 NO | — | 11c | — | $0 | PASS — above my 10c cap |
Total at risk: ~$9.39
Norris NO moved above my price cap, so I passed.
Payoff scenarios — payoff mechanics, not forecasts.
| Outcome | Combined payout | Approx. Baku P&L |
|---|---|---|
| Russell does not win; Leclerc outside Top 10 | $27.32 | +$17.93 |
| Russell does not win; Leclerc makes Top 10 | $23.32 | +$13.93 |
| Russell wins; Leclerc outside Top 10 | $4.00 | -$5.39 |
| Russell wins; Leclerc makes Top 10 What happened | $0.00 | -$9.39 |
Main risk: this portfolio is highly concentrated in Russell not winning. If Russell wins, the Leclerc NO position alone cannot cover the combined stake.
This was race 1 of my three-race allocation test. The test has not proven anything yet.
This is an experiment. Not investment advice.
3-race allocation test: wrap-up
| Race | Policy | Result |
|---|---|---|
| 1 Baku | A-1.3, concentrated contrarian NO | -$9.39 |
| 2 Sepang | A-1.4.x barbell | ~+$0.03 |
| 3 Singapore | A-1.4.1 barbell, higher actual entry prices | -$0.17 |
| Test total | mixed policy, not three identical tests | ~-$9.53 |
Most of the test loss came from Baku's concentrated Russell Winner NO position. The two barbell races together came to about -$0.14. Three small races don't show an edge, or the lack of one. This is an experiment, not a validated strategy.
Live ledger
| Race | New capital | Net P&L | Ending value |
|---|---|---|---|
| Dutch / Zandvoort | $5.0000 | -$1.0662 | $3.9338 |
| Italy / Monza | $5.0000 | -$0.8300 | $8.1038 |
| Spain / Madring | $5.0000 | -$0.6666 | $12.4372 |
| Azerbaijan / Baku | $5.0000 | -$9.3884 | $8.0488 |
| Malaysia / Sepang | $5.0000 | ~+$0.03 | ~$13.0788 |
| Singapore / Marina Bay | $5.0000 | -$0.17 | ~$17.9088 |
Totals (Quant): contributions $30.00 · value ~$17.91 · net P&L ~-$12.09 (-40.30% on contributions). These are not audited broker cash.
(Debit/Payout: Sepang ~$5.01 / $5.04 · Singapore $5.49 / $5.32.)
Latest Short
No Project F1 Short published yet. Archive will list each race Short here when live.
Latest analysis
Download latest-analysis.mdProject F1 — Singapore Grand Prix 2026 | Post-race results and director handoff
Event: Singapore Grand Prix, Marina Bay — 11 October 2026
Test: Race 3 of 3 completed (mixed allocation policies: Baku A-1.3; Sepang and Singapore A-1.4.x barbell)
Prediction model: Price-blind v1.3.0 team-aware, 500,000 simulations, master seed 20261010
Race and two purchased Top-10 outcomes: Confirmed. All four economic results can be determined from the race winner and purchased payout terms.
1. Official race classification (publishable)
| Finish | Driver | Team |
|---|---|---|
| 1 | Max Verstappen | Red Bull Racing |
| 2 | Charles Leclerc | Ferrari |
| 3 | Kimi Antonelli | Mercedes |
| 4 | Lando Norris | McLaren |
| 5 | Oscar Piastri | McLaren |
| 6 | Lewis Hamilton | Ferrari |
| 7 | Liam Lawson | Racing Bulls |
| 8 | Nico Hülkenberg | Audi |
| 9 | George Russell | Mercedes |
| 10 | Isack Hadjar | Red Bull Racing |
Official classification: https://www.formula1.com/en/results/2026/races/1296/singapore/race-result
Verstappen won from pole, 7.335 seconds ahead of Leclerc, with Antonelli third. The full top-ten order above comes from the official race classification. The race report notes a late VSC/strategy decision contributed to Hamilton falling to sixth; do not assert that the model forecast this incident. Race report: https://www.formula1.com/en/latest/article/verstappen-seals-commanding-singapore-grand-prix-victory-from-leclerc-and-antonelli.49wDzysiVqi5cloGjle40A
6. Public-facing editorial summary
Suggested headline: “Our model picked the winner. Our safer picks paid. The portfolio still lost a little.”
Public claims supported: The model ranked Verstappen as its most likely winner after 500,000 Singapore simulations; he won. The actual podium was Verstappen, Leclerc and Antonelli. The barbell portfolio held Verstappen and Antonelli Top-10 YES (both paid) alongside Hamilton and Leclerc Winner YES (both lost). It was a slight net loss, not a profit. The experiment is three races into a completed, mixed-policy allocation test, not a validated winning strategy.
Suggested card sequence:
- SINGAPORE — RACE RESULTS: Verstappen P1 · Leclerc P2 · Antonelli P3.
- MODEL CHECK: 500,000 simulations · Model's top win pick: Verstappen · Outcome: P1.
- OUR BARBELL: Top-10 core: Verstappen + Antonelli · Winner upside: Hamilton + Leclerc.
- WHAT PAID: Both core Top-10 contracts ✔ · Both Winner bets ✕.
- THE LESSON: Accurate finishing picks do not guarantee portfolio profit; entry price matters.
- TEST REVIEW: Three races completed; mixed allocation methods; evaluating what worked, what didn't, and execution speed.
Optional first-person narration (proposed script, not a historical quote):
“In Singapore, my F1 model ranked Verstappen as the most likely winner—and he won. Leclerc and Antonelli completed the podium. My barbell strategy got both higher-probability top-ten picks right, but the two winner long shots lost. And because of the prices I actually paid, the overall portfolio finished slightly down. That is the point of Project F1: being right about a race isn't the same as making money on a prediction market. Three races into the allocation test, I'm reviewing both the model and how fast I can act on its predictions.”
Editorial source links: official classification and race report above.
Note: Economic totals use cent-level confirmed holdings figures where provided; sub-cent exact cash amounts are intentionally not invented.
What’s going on
6 races completed (Dutch, Italy, Spain, Azerbaijan, Malaysia, Singapore). Contributions: $30.00. Current value: ~$17.91. Cumulative net: ~-$12.09. Last race: Singapore, -$0.17. As of 13 October 2026.
Disclaimer
Results are historical and experimental, not investment advice. Prediction markets involve risk, including loss. Simulated results are not funded returns.