What has been checked, and what has not

The project includes repeated scenario tests and logical consistency checks. It does not yet claim professional validation or full calibration against national collision-history datasets.

View results Assumptions register

Reliability at a glance

Checked, not checked, and next validation

Page Purpose

Reliability at a glance

This summary separates working checks from the validation work still required before stronger scientific claims.

Checked: Internal model direction, scenario response, graph wording, export structure, and reviewed saved records.
Not checked: Professional engineering validation, field inspection, traffic exposure calibration, and national collision-history calibration.
Next validation: Compare many modelled bins against observed RSA collision density and severity-weighted context.
Boundary: RSA context supports review and calibration planning. It does not prove causation or certify road safety.

Validation status

Current reliability position

Validation area	Current status	What it means
Internal consistency	Done / ongoing	Core directions such as curvature, radius, speed, friction, and stopping distance behave as expected.
Repeated scenario testing	Checked in App V2	Controlled selected-assumption changes produce explainable model response.
Multiple road forms	Checked in reviewed cases	Saved records include different road classes and geometry contexts.
Graph and percentile context	Partly checked	Graph interpretation is documented, but percentile context is not stored in every public case.
Export evidence trail	Supported	Case JSON, CSV, GeoJSON, graph data, and local saved cases can document outputs.
RSA observed context	Available in App V2	RSA Recorded Collisions can be compared near selected roads as historical context, not proof of model correctness.
Model-to-observed comparison	Available for selected-road bins	Nearby RSA records can be counted against selected-road bins with careful wording and minimum-data warnings.
Collision-history calibration	Not completed	The model is not calibrated against full official collision datasets.
Field inspection	Not completed	Surface, visibility, drainage, signage, and defects require real-world inspection.
Professional engineering review	Planned separately	Operational use would require expert review and formal validation.

Observed collision context

How RSA Recorded Collisions are used

Historical records

RSA Recorded Collisions are historical recorded collision records from RSA source data. They are presented as observed context and are separate from Road Risk model outputs.

Selected-road comparison

App V2 can count nearby RSA records within an explicit radius and compare them with selected-road bins. This helps review spatial alignment and face validity without claiming causation.

Boundary

A local match is useful evidence that the integration is working, but it is not proof that the model predicts collisions or that any road has an official safety rating.

Internal checks

Consistency checks expected from the physics

Sharper curvature should generally increase comparative output under the same scenario.
Smaller radius should reduce the friction-limited safe-speed estimate.
Higher speed should increase stopping distance non-linearly.
Lower effective friction should reduce safe speed and increase braking distance.
Percentiles should be interpreted only within the sampled comparison set.
Fallback-heavy cases should carry lower interpretation confidence.

What has not been validated

No operational certification yet

What has been checked

The project has checked internal model direction, scenario response, graph interpretation wording, exported case structure, and reviewed saved records.

What has not been validated

Road Risk has not completed collision-history calibration, field inspection, professional engineering review, or operational certification.

Stronger validation path

Future calibration and reliability work

Collision data

Scale aggregate comparison

Test whether higher model outputs align with trusted collision-history datasets across many roads once ethical and data-quality conditions are met.

Known sites

Check known high-risk bends

Compare selected outputs with locations already identified by professional or public-sector processes.

Engineering review

Compare against audit expectations

Ask road-safety or transport professionals whether model mechanisms match field-audit reasoning.

Field data

Verify surface and visibility

Inspect whether OSM tags, surface assumptions, sightlines, lighting, and barriers match real road conditions.

Sensitivity testing

Run assumption ranges

Run controlled sensitivity analysis across friction, speed, reaction time, visibility, vehicle, and traffic-proxy settings.