Works in preview. Breaks live.

1. Write down both addresses.

   Open preview and the public app in separate tabs. Record the exact failing action and which tab fails. Confirm you are comparing the expected version, not an older browser tab or a different project with a similar name.

2. Compare the requests.

   In each tab open Inspect, then Network. Run the action once and select the request it triggers. Compare its destination host, path and status. Read the failed Response, with private values removed before sharing.

3. Read the production build result.

   On Vercel, open the project’s Deployments list, select the relevant deployment and expand Building in its details. On Netlify, open Deploys, select the deploy and read its deploy log. Compare the command, revision and first failure with what preview actually runs.

4. Check the production configuration.

   Open the host’s environment settings and compare required variable names and production scope with the code. Do not print secret values into logs. Note any variable that exists only in preview, then use the environment guide for the appropriate host.

WORKSThe same test journey completes against the intended live service.

BROKENProduction’s build, destination or response differs in a way that prevents the journey.

Ranked in the order to investigate, not by claimed frequency.

1. Live runs another revision or build command.

   Select the intended source and production build configuration. Repair the first build failure rather than changing unrelated UI code. Open the resulting deployment address and repeat the exact failing journey.

2. Production points at the wrong service.

   Correct the production destination and required configuration. Keep test and live projects deliberately separated; copying all preview settings blindly may route real users into test data.

3. The published hosting type cannot run the backend.

   Confirm whether the app needs a running server or only static files. Deploy the server portion to an appropriate runtime and make the client call that deployed endpoint.

• A custom domain can reach a different deployment from the host’s deployment URL. Compare both before assuming the build itself is wrong.
• A homepage loading does not prove an inner route or server action works. Reproduce the exact failing path and action rather than stopping at the first screen.

When to stop and hand it over

Hand over if the build and live requests disagree about which project is running, or if the repair requires changing hosting architecture. Send both URLs, the deployment identifier and the first relevant log excerpt after removing secrets.