Competition review route

5-minute routePublic reviewSign-in optional

Judge Route: Review Road Risk in 5 Minutes

Use this page to understand what the project tests, what evidence exists now, what the live app can demonstrate, and what the project does not claim.

Open Live Demo View Evidence

Fast route

Review sequence

This is the shortest public path through the project. Account and Dashboard are private-workspace features and are not required for judging the public research site.

Page Purpose

Official five-minute review pathway

Purpose: Provide the shortest public path for reviewing the research question, method, evidence, live demo, and authorship.
Read first: Start with this page, then use Method and Results before opening App V2.
Demo handoff: App V2 is the live demonstration; Account and Dashboard are not required for public judging.
Limit: This route reviews model behaviour and evidence framing. It does not prove real-world collision risk.

Read this page

Confirm the research question, hypothesis, variables, limits, and judging route.

Open Method

Check that data, derived geometry, assumptions, outputs, and context are separated.

Open Results

Review Phase 1 model-behaviour evidence from the committed public case dataset.

Open App V2

Select a road, change one assumption, inspect Maths, and export evidence if needed.

Read Authorship

Check what was built, what tools were used, and what help still needs named attribution.

Research frame

Question, hypothesis, and variables

Research question

Can public geometry support a prospective signal?

Road Risk asks whether public road geometry, physics-informed calculations, and explicit scenario assumptions can produce transparent comparative outputs before formal collision-history calibration.

Hypothesis

Higher geometric and scenario demand should raise model output

Under the same assumptions, road sections with tighter curvature, greater speed demand, longer stopping distance, and weaker data confidence should produce higher comparative outputs. This is a model-behaviour hypothesis, not proof of real collision risk.

Changed variables

Speed, weather, surface, visibility, lighting, vehicle profile, fatigue, overspeed, and selected road geometry.

Controlled conditions

The same selected road can be tested while changing one scenario setting at a time.

Measured outputs

Annual model output, local bin output, safe-speed estimate, stopping distance, confidence, and graph position.

Context layer

RSA records are historical observed context where configured. They are not Road Risk model output.

What to check

The project evidence is strong enough for review, not final validation

Phase 1 evidence is inspectable

The public Results page contains reviewed model-output records that can be filtered, charted, and opened without signing in.

Model behaviour can be tested live

App V2 lets a reviewer select a road and change assumptions deliberately, rather than reading a static claim.

The boundary is explicit

The site repeatedly separates modelled outputs, historical collision context, private dashboard records, and official safety decisions.

Live demo script

A practical demonstration path

Demo

Run the live app

Open the current map workspace, select a road, and inspect the selected-road output.

Open App V2

Method

Explain the model

Show formula traces, assumptions, data confidence, and interpretation boundaries.

Formula Chain

Evidence

Show evidence

Review public cases and explain what Phase 1 model-behaviour evidence can and cannot show.

View Results

Authorship

Check contribution

Review the build summary, tools used, data sources, and help-received boundary.

Read Authorship

Limits

What this project does not claim

Scientific boundary

Road Risk is a transparent comparison and evidence-building tool. It should be judged on whether the model chain is inspectable, the assumptions are visible, and the outputs are responsibly framed.

Road Risk is not an official road-safety rating.
The model does not calculate observed crash probability for a named road.
RSA records are historical context and do not prove the model is correct.
Collision-history calibration, field inspection, traffic exposure data, and expert review remain necessary for stronger validation.

Next work

What I would do next

Calibration

Compare model outputs against larger RSA observed-collision samples across many roads and local bins.

Demo mode

Create one safe App V2 judging demo with a known route, fallback screenshots, and a reset state.

Authorship detail

Before final submission, add exact named teacher, mentor, and external feedback details where required by the competition rules.