> ## Documentation Index
> Fetch the complete documentation index at: https://offthepace.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Which circuits suit each driver's style?

> A heatmap of how much faster or slower each driver is at a specific circuit, relative to their own season average, after removing car and field effects.

Some drivers seem to find an extra tenth at certain tracks year after year. This heatmap measures that: for each driver and circuit, how much faster or slower they tend to be there relative to **their own** season-average pace, after constructor and field effects are removed. Negative (green) means faster than the driver's average at that circuit; positive (red) means slower.

## What it shows

Each cell is one driver × circuit affinity:

* **Shrunk affinity**: a Bayesian posterior mean that pulls the raw observed circuit delta toward the driver's global mean, using 5 virtual races as the prior weight. Drivers with fewer observations are shrunk more aggressively toward neutral.
* **Sign convention**: negative = faster than the driver's own average at that circuit (green); positive = slower (red) the same convention as the driver skill residual.
* **Confidence**: the fraction of the posterior coming from data versus prior, `n_obs / (n_obs + 5)`. Values below 0.17 (one race) are prior-dominated.

Drivers are sorted fastest-to-slowest by median affinity across circuits; circuits are sorted alphabetically.

<Note>
  Cells are shown only when `n_obs ≥ 2` (two or more races at the circuit). Single-race visits are excluded as unreliable, so empty cells mean too few visits, not zero affinity.
</Note>

## How to use it

1. **Select a season scope** from the filter bar; the heatmap reloads.
2. **Scan a driver's row** for their strong (green) and weak (red) circuits.
3. **Scan a circuit's column** to see which drivers historically over-deliver there.
4. **Check confidence before concluding** a strong colour on a low-confidence cell is mostly prior, not signal.

## Reading the signal correctly

Affinity is **within-driver**: it compares a driver to themselves, so it is not a pace ranking. A slow-car driver can show strong positive affinity at a circuit that suits their style, independent of where they finished.

<Warning>
  Affinity and the era ratings share an underlying skill residual; do not stack them as if independent. A circuit where a whole era's data is thin (few events) will show wide, prior-dominated cells read the confidence column. See the [decomposition limitations](/decomposition/limitations).
</Warning>

## Data source

The chart queries `int_driver_circuit_affinity` via DuckDB-Wasm in the browser.

| Column              | Meaning                                                                        |
| ------------------- | ------------------------------------------------------------------------------ |
| `shrunk_affinity_s` | Bayesian posterior circuit delta vs the driver's own average (the cell colour) |
| `n_obs`             | Races the driver ran at the circuit; gates display at `≥ 2`                    |
| `confidence`        | Data fraction of the estimate, `n_obs / (n_obs + 5)`                           |

The data window covers **2018–2024**.

For the full model definition, see the [`int_driver_circuit_affinity` reference](/reference/models/int/int_driver_circuit_affinity).