> ## 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 drives were better than the result showed?

> Drivers whose ghost-projected finish beat their actual result strong drives masked by car pace, strategy, or circumstance.

A points-less afternoon can hide a great drive. Hidden performance surfaces those: it compares each driver's ghost-projected finish what the model expected given their pace in a chosen host car against where they actually finished. A driver who was projected higher than they finished was, in pace terms, under-rewarded by the result.

## What it shows

The table ranks drivers by the gap between projection and result:

* **Ghost-projected position (`fct_ghost_race_finish`)**: each driver's lap times recombined with a chosen host constructor's car pace, then re-ranked by predicted cumulative race time.
* **Delta**: projected minus actual finish. A *negative* delta means the model projected a better finish than actually occurred the driver was under-rewarded.
* **Confidence**: how well the host constructor's pace is estimated from panel data; more races with that team means higher confidence.

<Note>
  Only scenarios where `avg_recombination_confidence ≥ 0.3` are included, and short runs with too few laps for a reliable estimate are excluded.
</Note>

## How to use it

1. **Pick a host constructor** from the filter bar; projections re-score against that car.
2. **Sort by delta** to surface the most under-rewarded drives.
3. **Read confidence** before trusting a large delta a low-confidence host pace makes the projection noisier.
4. **Use the identity check** as a sanity test: when the host equals the driver's own team, projected should equal actual.

## Reading the signal correctly

The delta is **pace versus result**, not an injustice ledger strategy, safety cars, and incidents legitimately move finishing positions in ways a pace projection cannot see. A negative delta says "faster than the result", not "robbed".

<Warning>
  When the host constructor equals the driver's own team, predicted finish should equal actual finish any non-zero delta in that case signals a data issue, not hidden performance. Treat it as a diagnostic, not a finding.
</Warning>

## Data source

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

| Column                         | Meaning                                            |
| ------------------------------ | -------------------------------------------------- |
| `projected_finish_pos`         | Ghost-projected finishing position in the host car |
| `actual_finish_pos`            | The real finishing position                        |
| `finish_delta`                 | Projected minus actual; negative = under-rewarded  |
| `avg_recombination_confidence` | Panel-data confidence; gated at `≥ 0.3`            |

The data window covers **2018–2024**.

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