> ## 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.

# A driver's track record in equal machinery

> A per-circuit, per-era leaderboard ranking drivers by car-removed pace, so the board reads as driver skill at that track rather than whose car was quickest.

What if every driver who ever raced this circuit had been in the same car? Ghost standings answers that. For one circuit and one regulation era, every driver is ranked by their **car-removed pace**, so the board reads as driver skill at that track not whose machinery was quickest. Negative means faster than the field once the car is removed.

## What it shows

The leaderboard is built from the driver-skill residual of the seven-term identity:

* **Equal car**: each lap is decomposed into fuel, tyre compound, rubber, ambient, *constructor*, and driver-skill components. The board uses only the driver-skill residual the part left once car and conditions are removed so a slow car no longer hides a fast driver. Negative = faster than the field.
* **Split by era, not blended**: records are computed separately for 2018–2021 (last-gen aero) and 2022–2024 (ground-effect), split on the 2022 regulation change. Pace is only ever compared within one car era.
* **No single year**: every race a driver ran at the circuit within the era is pooled, so this is a multi-season record.
* **Shrunk toward the driver's own average**: a one-race sample is regularised toward the driver's era-wide mean (Bayesian, prior = 5 virtual races). The **Conf.** column is the data fraction `n / (n + 5)`; values below 0.3 (amber) are prior-dominated.
* **95% CI**: overlapping intervals mean two drivers are statistically tied at that track.
* **Min. races**: filters thin samples; the gap-to-leader is measured against the fastest driver shown.

<Note>
  This is a statistical reconstruction of equal-car pace, not a physics simulation. Read the ranking as expected-value estimates, with the confidence and CI columns as the uncertainty.
</Note>

## How to use it

1. **Select a circuit and era** from the filter bar; the leaderboard loads.
2. **Read the ranking** as track-specific driver skill, car removed.
3. **Check the Conf. column** amber (\< 0.3) rows are mostly prior and mean "not enough races yet".
4. **Compare CIs before ranking two rows** overlapping intervals are a tie.

## Reading the signal correctly

The board is **within one era and one circuit** by design never average a driver across the 2022 boundary, and don't read it as an all-time ranking. The skill residual is what survives car removal, so a midfield driver can top a circuit where they consistently over-delivered.

<Warning>
  Low-confidence (amber) rows can post extreme positions on a single hot weekend before shrinkage fully reins them in. Weight the confidence and CI columns, not just the headline order. See the [seven-term identity](/decomposition/seven-term-identity) for how the residual is isolated.
</Warning>

## Data source

The chart queries `int_driver_circuit_era_affinity`, built on the driver-skill residual of `fct_lap_residuals`, via DuckDB-Wasm in the browser.

| Column                      | Meaning                                                              |
| --------------------------- | -------------------------------------------------------------------- |
| `shrunk_skill_residual_s`   | Car-removed pace at the circuit, shrunk toward the driver's era mean |
| `confidence`                | Data fraction `n / (n + 5)`; amber below 0.3                         |
| `ci_lower_s` / `ci_upper_s` | 95% credible interval                                                |
| `regulation_era`            | 2018–2021 vs 2022–2024 split                                         |

The data window covers **2018–2024**, split at the 2022 regulation change.

For the full model definitions, see the [`int_driver_circuit_era_affinity` reference](/reference/models/int/int_driver_circuit_era_affinity) and the [`fct_lap_residuals` reference](/reference/models/fct/fct_lap_residuals).