cairnIN PRACTICE

A BROWSER TEST, FROM START TO FINISH

One flow.
Every release.

You ship. The checkout still needs checking.

Describe the task once. cairn discovers a test, saves it, and replays it without a model. Follow a real run below — including what happens when things break.

Follow the checkout ↓
THE TASK

“Order one Field notebook and verify the order succeeded.”

SAMPLE APP / FIELDNOTES

A test that has somewhere to go.

RECORDED EXECUTION · NOT LIVE

Loading the execution record…

THE PART THAT MATTERS

A changed button is a test repair.
A failed order is still a failed order.

This example changes the order control from a button to a link. Then it makes the order API return 500, even though the screen says “complete.” The test checks the request, not just the destination.

TAKE IT FOR A WALK

Run the same story yourself.

Node 22.12+, Chrome, and an authenticated model backend.
Recording uses model calls; browsing this page uses none.

Download runnable source ↓Read the runner ↗Inspect the execution record ↗ 
tar -xzf source.tar.gz
cd order-demo
npm install
CAIRN_LLM_BACKEND=codex DEMO_MODEL=gpt-5.6-sol npm run record
npm start

One measured example, not a general repair success rate. Each stage starts with a fresh browser. Model costs are not shown when the backend does not report them. This demo is ready for usability testing; no external-user outcomes are claimed.