Model and race simulation method

How the 2026 Formula 1 model handles pace, strategy, tire wear, deployment, and race control.

The current build aims for a believable 2026 Formula 1 weekend model without pretending to be a lap-perfect physics simulator.

1. 2026 season catalog

The app now uses the real 2026 Formula 1 teams, drivers, Grands Prix, and Sprint weekends. The performance and circuit priors layered on top of that catalog are estimated for the simulator and should be read as modeled inputs, not official timing data.

2. Pace prior

A compact PyTorch MLP predicts a baseline pace prior from driver and circuit features. It does not predict the finishing order on its own. It only supplies the starting pace signal for the wider race simulation.

3. Deterministic race logic

The explicit model handles qualifying weight, tire degradation, pit loss, fuel sensitivity, reliability pressure, and 2026-specific deployment and active-aero framing. Those assumptions stay visible on purpose.

4. Race control and Monte Carlo

Weather shifts, yellow flags, VSCs, safety cars, red flags, incidents, and DNFs are sampled repeatedly. The result is a probability distribution for the configured weekend, not a single hard forecast.

5. Historical backtesting and calibration

The simulator now includes an official-source-backed historical pipeline, normalized weekend schema, and backtesting workflow used to tune circuit leverage, strategy pressure, and race-control assumptions against a focused set of real Grands Prix.

6. Trust and calibration layer

Each scenario now carries an explicit trust summary: confidence tier, historical support tier, calibration depth, data grounding, and volatility. Those trust labels are derived from the current historical support basket, backtest coverage, and scenario complexity rather than from marketing-style certainty scores.

7. Provenance separation

The product keeps official source data, normalized historical datasets, modeled seed priors, calibrated parameters, and live user assumptions separate on purpose. The simulator does not present modeled assumptions as if they were official FIA / Formula 1 facts.

How provenance is separated

The simulator distinguishes source-backed evidence, normalized project data, calibrated layers, and live scenario assumptions instead of blending them together.

Official source data

Formula 1 result pages, starting grids / qualifying pages, and pit-stop summaries used in the historical pipeline.

Normalized historical data

Project-shaped historical weekend files with explicit provenance tags for derived weather and neutralization markers.

Modeled priors

2026 pace priors, weather/disruption assumptions, and user-entered setup levers used to simulate scenarios that are not directly observed.

Calibrated layers

Circuit leverage, overtaking, strategy pressure, and event logic tuned against the current historical subset and backtest reports.

How to read confidence

Confidence is a calibrated product signal, not an official probability of truth. It reflects support depth, volatility, and similarity to the current historical basket.

High confidence: well-supported, lower-chaos scenario families where the simulator behaves closest to a calibrated probability map.

Moderate confidence: historically anchored enough to trust the broad race shape, but still materially sensitive to race-state evolution.

Experimental / Low confidence: thinner circuit support, heavier chaos, wet crossover pressure, or scenario complexity beyond the strongest calibration set.

What the current build handles well

The model is strongest when it is used as a scenario tool for race weekend tradeoffs rather than as a claim of exact lap-by-lap truth.

Circuit-specific behavior for overtaking, qualifying weight, tire stress, safety-car risk, and deployment sensitivity

Strategy recommendations tied to weather, track position, degradation, and race-control assumptions

Compare Mode for structured A/B scenario decisions instead of only single-board reads

Driver explanations built from actual model signals instead of generic summary text

Points-aware weekend outputs, including expected race points and constructors contribution

Scenario trust summaries that separate historically grounded baselines from more experimental weather- or chaos-heavy reads

What is still simplified

This is a structured Grand Prix simulator, not a full race-engineering stack.

Race resolution is event-aware and stint-aware rather than full lap-by-lap

2026 deployment and active-aero behavior are modeled as scenario levers, not detailed control laws

Qualifying is represented through leverage and pace rather than a separate session simulator

Team and driver priors are estimated inputs, even though the 2026 season entities are real

Historical support is still a small initial basket rather than a broad multi-season calibration set

Next realism steps

The current architecture is already shaped for deeper Formula 1 realism without a rewrite.

Qualifying simulation and Q3 probability outputs

Broader historical coverage across more circuits and weather families

More detailed Sprint-weekend handling

Richer FIA race-control and lap-window weather ingestion