Off The Pace app: browser-native F1 analytics

A zero-server React app powered by DuckDB-Wasm and ONNX. Parquet files load in your browser; every SQL query runs locally in under 10 ms.

Off The Pace is a browser-native F1 analytics application deployed at off-the-pace.web.app. There is no compute server when you open the app, your browser downloads Parquet files, loads them into DuckDB-Wasm running in WebAssembly, and executes SQL queries locally in under 10 ms. The ML models (XGBoost quantile regressors, cliff classifiers, stint-life regressors) are bundled as ONNX and scored in-browser. Nothing you interact with touches a backend at runtime.

For a reviewer:

• Decision: a zero-server React app; DuckDB-Wasm and ONNX Runtime Web run every query and every inference on the client.
• Trade-off: a larger initial download and no row-level authorization, in exchange for free, offline-capable serving.
• Proof: Playwright E2E drives the live production bundle and a bundle-size budget gates every build.

Architecture

Data loads in your browser no server processes your queries. Three layers run entirely inside your browser tab:

Parquet files  ·  ONNX model bundles
      ↓
DuckDB-Wasm  (WebAssembly, sub-10 ms SQL)
      ↓
Charts, simulators, and query editor

The Parquet files are the same gold-mart tables produced by the dbt pipeline fct_lap_residuals, fct_driver_skill_features, fct_cliff_prediction_features, dim_compounds_season, and a handful of dimension tables. A manifest resolves every table path at runtime so you always query the latest published build.

Available visualizations

The app is organized into content pillars. The following are live and available today:

Pillar	Path	What you get
Ghost Car	/ghost-car	Compare simulated optimal laps vs actual laps
Lap Decomposition	/lap-decomposition	Decompose any lap into its eight additive causes
Tyre Strategy	/tyre-strategy	Dial a compound and conditions to watch the degradation models
Aero & Conditions	/aero	Analyze track temperature and dirty air impacts
Drivers	/drivers	Driver consistency charts and era-normalized skill ratings
Constructors	/constructors	Team structural pace contribution and trends
The Machine	/ml	ML model metrics, blind-test scoreboard, and the degradation simulator
Query Lab	/query	A full SQL editor run any query against the gold mart in-browser
Data Quality	/data-quality	Data pipeline invariants and telemetry reliability checks

The following pillars are deployed as placeholders and are coming soon: Races, Race Craft, Energy & Telemetry.

How to use the app

1. Navigate to off-the-pace.web.app.
2. Use the filter bar at the top of every page to select a season and, where relevant, a specific race.
3. Pick a pillar from the sidebar to explore the visualization you want.
4. The engine status indicator in the top bar shows when DuckDB-Wasm has finished loading queries run as soon as it turns green.

Explore by pillar

Ghost Car

Driver skill in equal machinery ghost recombination, counterfactual championships, and hidden strong drives.

Lap Decomposition

The seven-term identity at every scale per-lap waterfall, full-race totals, and per-sector splits.

Tyre & Strategy

Survival curves, degradation timelines, pit-strategy Gantts, and recovery forecasts.

Aero & Conditions

Dirty-air cost and lap maps, track rubber buildup, and the field pace baseline.

Drivers

Nine angles on driver skill ratings, circuit affinity, wet-weather edge, corner technique, and teammate duels.

Constructors

Car pace isolated from the driver structural pace and the circuit interaction heatmap.

The Machine

Live ONNX-scored degradation simulator, model metrics dashboard, and held-out blind-test scoreboard.

Data & Validation

Query the gold marts directly in SQL, and audit per-race lap exclusions from the pipeline.