A raw model value is easier to misread than a contextual one. Road Risk uses distribution insights so a selected road can be compared with sampled roads in the surrounding network.
A selected road's Annualised Comparative Model Output is informative, but it becomes more useful when compared with a sampled distribution. If most nearby roads are straight and one road has an elevated curvature-driven output, the percentile view makes that contrast visible.
Graphs are not decoration. They help answer whether the selected road is typical, elevated, unusual, or part of a high-risk tail under the current scenario assumptions.
Percentiles are comparative within the sampled road context. They do not prove that a real collision will happen; they show how one modelled road compares with the sampled modelled network.
If a road sits at the 87th percentile, it means it is higher than 87% of the sampled comparison set under the current assumptions.
That does not make it universally dangerous. It means the selected road is elevated relative to the roads sampled in that view and scenario.
Percentile context depends on the selected area, sample radius, visible road mix, public-data completeness, and the scenario assumptions active at the time.
Shows the shape of the sampled risk distribution and whether the selected road sits in a dense or unusual region.
The empirical cumulative view makes percentile interpretation explicit and easier to explain.
Robust distribution views show median, quartiles, tails, and whether high-risk values are rare or common.
Tail-focused views help identify whether risk is broadly distributed or concentrated in a smaller set of segments.
Upper-percentile markers help separate ordinary variation from the highest tail of sampled model outputs.
Before/after distributions can show how rain, fog, speed, fatigue, or vehicle profile shifts the sampled network.
| Graph Type | What It Shows | When to Use It | What Not to Conclude |
|---|---|---|---|
| Histogram / KDE | Distribution shape and density of sampled model outputs. | Use when judging whether the selected road sits in a common or unusual part of the sample. | It is not a national distribution unless the sample is national. |
| ECDF / CDF | Cumulative position and percentile interpretation. | Use when explaining how much of the sampled set lies below the selected output. | A percentile is not a universal safety rank. |
| Box Plot / Violin | Median, quartiles, spread, and tail shape. | Use when comparing stable and variable samples. | It does not show the cause of each road's output on its own. |
| Exceedance / Upper Tail | How many samples exceed a chosen output or threshold. | Use when identifying whether elevated outputs are rare or common in the current context. | Exceedance is model-contextual, not observed collision frequency. |
| Contribution / Component Breakdown | Which model factors contribute most to a displayed output. | Use when explaining whether geometry, speed, friction, or scenario assumptions dominate. | Component bars should not be treated as independent crash causes. |
| Compare Mode | Before/after distributions under changed assumptions. | Use when testing rain, fog, fatigue, overspeed, or vehicle-profile sensitivity. | Scenario shifts are assumption tests, not measured environmental records. |
Graph data, risk cards, maths output, exports, and map overlays should be based on the same selected-road state.
For that reason, the public site presents the graphs as part of the evidence trail rather than as isolated analytics.
The chosen road can be placed on the distribution to show relative position.
Weather, speed, vehicle type, and behaviour assumptions should be interpreted alongside the distribution view.
Distribution-data exports make the graph reproducible and easier to audit after the browser session.
Comparing dry and wet assumptions can show whether a road remains typical or moves toward the upper tail.
Upper-tail thresholds help identify sampled roads that are unusually high within the current comparison set.
A wider distribution can mean the sampled roads respond very differently to the same scenario assumptions.
Percentiles depend on the selected map area, sampled road mix, radius/query settings, and available public data.
Changing weather, driver, speed, or vehicle assumptions changes the comparison set and selected-road interpretation.
A high percentile means high relative output within the sampled model context, not proven real-world danger.
Open Distribution Insights after selecting a road to compare the selected segment with sampled surrounding roads.