flowchart LR A[Past job records] --> B[Patterns] B --> C[Made-up calls] C --> D[Rules choose a vehicle] D --> E[Results]
Overview
NoteIn brief
- This is a computer model that imitates how Devon Air Ambulance’s helicopters and critical care cars respond to emergency calls. Planners use it to try out changes safely on a computer.
- It does not send real helicopters, it does not handle real 999 calls, and it makes no decisions about any patient.
- It uses made-up calls, built from patterns in two years of real job records (2023 and 2024).
- Some of its results still differ from the real figures. See “Known issues” below.
What this is
Imagine a very detailed board game. Each day, pretend emergency calls arrive. The game decides which helicopter or car, if any, goes to each one. It then writes down what happened.
That is what this model does, on a computer. Experts call it a simulation.
The calls are made up. But they follow the same patterns as real calls: how many arrive each day, at what times, what kind they are, and how long each stage of a job takes.
Planners can then ask “what if?” questions. What if shifts were longer? What if there were an extra car? What if a different aircraft were used? The model shows what might happen, before anyone tries it for real.
What it produces
- Estimates of how many calls would be missed because no suitable helicopter or car was free.
- How busy each vehicle and crew would be.
- A side-by-side look at how these figures compare with what really happened in 2023 and 2024.
You see these in a web app. You can also download a report, if the right free publishing tool is installed.
The journey from records to results
What it does shows the longer version. The model has eight working parts, all explained in the glossary: the historical data preparer, the pattern sampler, the call simulator, the fleet and rota model, the dispatch rules, the run manager, the results and comparison tool, and the web app.
Who made it, and how to try it
The authors are Sammi Rosser, Richard Pilbery and Hannah Trebilcock. The model’s instructions are published openly, so anyone can read them. There is also a public demonstration copy that you can try in a web browser. An outside company hosts it, and it does fewer runs, so its results are rougher.
If you have questions, the project keeps a public list of known problems and questions, called the issue list. It is on the project’s page on a code-sharing website. Anyone can read it or add to it.
Known issues
Some results still differ from the real figures. Assumptions and limits gives the plain-language list. Open issues change over time, so the live list may differ from that summary.
How to use this guide
| Page | The question it answers |
|---|---|
| What it does | What problem does it solve, and what does it produce? |
| Your data | What information does it use, and where does it go? |
| Reading results | How should you read the results, and what is easy to get wrong? |
| Assumptions and limits | What does it take for granted, and where can it be wrong? |
| Glossary | What do the terms mean? |
This guide describes what the code does and assumes. Where the code does not show something, the guide says so. It makes no claim about legal or regulatory matters.