2025 Season Backtest
How the model would have performed picking teams for every 2025 race using only practice data available before qualifying.
| Race | Pts | Best | Capture | Xfers | Team |
|---|
Calibration Curve
Average actual fantasy points earned per predicted finishing position, learned from 2025 data.
How It Works
Practice Analysis
All three practice sessions are ingested in real time. Single-lap pace, long-run degradation, and tire compound are all captured and normalized to produce a true picture of each driver's weekend form.
Predictive Model
Session data is weighted and combined using parameters optimized across the full 2025 season. The model accounts for tire compounds, sprint weekend formats, and driver-specific tendencies that practice alone doesn't reveal.
Race Simulation
Thousands of race scenarios are simulated with qualifying variance, safety cars, pit strategies, overtakes, and DNFs. The result is a full probability distribution of fantasy points, not just a single guess.
Team Optimization
Every valid combination of 5 drivers and 2 constructors is evaluated against the $100M budget. Multiple strategies (balanced, safe, aggressive, value) are scored to match your risk preference.
2026 Fantasy Scoring
Drivers
| Race P1-P10 | 25, 18, 15, 12, 10, 8, 6, 4, 2, 1 |
| Quali P1-P10 | 10, 9, 8, 7, 6, 5, 4, 3, 2, 1 |
| Position gained | +1 per position |
| Overtake | +1 per overtake |
| Fastest lap | +10 |
| Driver of the Day | +10 |
| DNF / NC | -20 |
| DSQ | -20 |
| Sprint P1-P8 | 8, 7, 6, 5, 4, 3, 2, 1 |
| Sprint DNF | -10 |
| Sprint fastest lap | +5 |
| Quali NC / DSQ / no time | -5 |
Constructors
| Both drivers Q3 | +10 |
| Both drivers Q2 | +5 |
| One driver Q3 | +3 |
| One driver Q2 | +1 |
| Both eliminated Q1 | -1 |
| Driver DSQ (quali) | -5 per driver |
| Driver DSQ (race) | -20 per driver |
| Pit stop < 2.0s | +10 |
| Pit stop 2.0-2.19s | +10 |
| Pit stop 2.2-2.49s | +5 |
| Pit stop 2.5-2.99s | +2 |
| Fastest pit stop | +5 bonus |
| Record pit stop (<1.8s) | +15 bonus |
Team Management
| Budget | $100M |
| Roster | 5 drivers + 2 constructors |
| Free transfers | 2 per weekend |
| Extra transfer | -10 pts each |
| Carry forward | 1 unused (no compounding) |
| DRS Boost | Doubles 1 driver (weekly) |
Chips (6 total, each once/season)
| 3X Captain | Triple 1 driver's score |
| Autopilot | Auto-assign best DRS Boost |
| Final Fix | Replace 1 driver pre-race |
| Wildcard* | Unlimited free transfers |
| No Negative* | All negatives set to zero |
| Limitless* | No budget cap for 1 race |
* Unlocked after Race 1
2026 Grid
| Team | Driver 1 | Driver 2 |
|---|---|---|
| Red Bull | VER Verstappen | HAD Hadjar |
| Mercedes | RUS Russell | ANT Antonelli |
| McLaren | NOR Norris | PIA Piastri |
| Ferrari | LEC Leclerc | HAM Hamilton |
| Williams | SAI Sainz | ALB Albon |
| Alpine | GAS Gasly | COL Colapinto |
| Aston Martin | ALO Alonso | STR Stroll |
| Haas | OCO Ocon | BEA Bearman |
| Audi | HUL Hulkenberg | BOR Bortoleto |
| Racing Bulls | LAW Lawson | LIN Lindblad |
| Cadillac | BOT Bottas | PER Perez |
Model Changelog
- Environmental session weighting -- Temperature delta, night race, and calendar density modifiers adjust session weights per-circuit
- Split weight optimization -- Separate B2B and standalone weekend weights (0.45 divergence, +0.13 error improvement)
- Travel stress modeling -- Timezone disruption inflates Monte Carlo variance and DNF rates
- 2026 learning loop -- Completed races feed back with 2x weight; shrinkage k decays as data grows
- Backtest improvement -- 80.6% capture rate (up from 74.5%), 4,429 pts across 23 races
- Post-qualifying model -- 60/40 quali/practice blend; post-quali earned 4,482 pts vs practice-only 4,429 pts (+53)
- Qualifying calibration -- Separate quali-position-to-points lookup table learned from 2025
- Per-driver DNF rates -- Historical 2025 rates replace flat 5% assumption in Monte Carlo
- PU component tracking -- Grid penalty risk assessment from power unit usage data
- Empirical pit stop model -- Team-specific pit stop point estimates from historical distributions
- Circuit adjustments -- Overtake rates, safety car probability, and DRS zones active in estimates
- Chip usage tracking -- SQLite-backed tracking of which chips remain available
- Monte Carlo race simulation -- Full probability distributions instead of single-point estimates
- Tire degradation modeling -- Per-driver deg curves from practice stint data
- Multi-mode optimizer -- Balanced, Safe, Aggressive, and Value strategies
- Weather integration -- Race-day forecast with confidence adjustment
- Track-specific modeling -- Circuit characteristics feed the simulation
- Uncertainty bands -- Point ranges, podium probabilities, and risk categories
- Sprint weekend support -- All 23 races covered including 6 sprint events
- Compound normalization -- Tire-adjusted lap times from calibrated 2025 data
- Qualifying prediction -- Per-driver qualifying tendencies improve constructor scoring
- Transfer-aware backtest -- Realistic season management with carry-forward logic
- Initial release trained on full 2025 season
- Empirical calibration with driver bias correction
- Budget-constrained team optimizer
Known Limitations
- Weather divergence between practice and race reduces prediction confidence
- On-track incidents and safety cars are inherently unpredictable
- Sprint weekends have less practice data available, so confidence is lower
- Drivers who changed teams for 2026 may behave differently than historical data suggests
- The model improves as the 2026 season progresses and more real data is available