> ## 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.

# The Machine: ML transparency and live inference

> Three features that expose the tyre-degradation ML models a live ONNX-scored simulator, a held-out blind-test scoreboard, and a full evaluation dashboard so the model's behaviour is auditable, not a black box.

## What this pillar does

The Machine is where the XGBoost tyre-degradation models become interactive and auditable.
Three features cover the full transparency arc:

* **Understand it.** Model Metrics shows the evaluation dashboard written at training time:
  temporal holdout performance, per-cohort baselines, conformal calibration intervals, and
  feature importance from SHAP and permutation.
* **Test it.** Blind-Test Scoreboard scores the model on a held-out season it never saw
  during training the hardest honest test of whether degradation predictions generalise.
* **Play with it.** Degradation Simulator lets you dial compound, fuel load, and conditions
  and watch the ONNX-exported XGBoost quantile regressors score a full stint live in your
  browser, with cliff probability and stint-life estimates updating in under 5 ms.

The models are documented in full in the [ML Pipeline](/ml/pipeline) and
[Validation & Trust](/ml/validation) sections.

## Every feature in this pillar

<CardGroup cols={2}>
  <Card title="Degradation Simulator" icon="sliders" href="/app/degradation-simulator">
    Dial compound, fuel, and conditions to watch XGBoost models predict pace loss, cliff risk,
    and tyre life scored live in your browser via ONNX.
  </Card>

  <Card title="Model Metrics" icon="bar-chart-2" href="/app/model-metrics">
    The evaluation dashboard for the tyre-degradation models temporal holdout, per-cohort
    baselines, conformal calibration, and SHAP-vs-permutation feature importance.
  </Card>

  <Card title="Blind-Test Scoreboard" icon="clipboard-check" href="/app/blind-test-scoreboard">
    The degradation models scored on a held-out season predicted-vs-actual scatter, interval
    coverage, and a cliff-class confusion matrix, all read from the model card.
  </Card>
</CardGroup>