Sample App Failures Are How SDK Evaluations End

Evaluators judge you by the repository you told them to clone. If it fails, they do not distinguish 'sample' from 'SDK.' They conclude the SDK fails. Your marketing site never hears about it.

Run the sample in CI on a clean machine, not only on the laptop that wrote it. Pin the CLI version. Fail the build when the happy path throws.

An empty catch around the first API call is how a 401 becomes a blank page. Show the code, the message, and the doc URL in the UI and the server log. Then emit the probe from that same catch, and rethrow or exit non-zero so CI sees it too.

If the sample needs Docker, say so in the first screen of the README, and map container exits to the codes from your CLI. A stack that mentions your framework and not the OOM is a missed diagnosis.

People change the sample immediately. They point it at staging, add a header, or skip a step. The stack shows your function. The sentence 'I was trying to use my own tenant' shows the gap in the README.

Ask that once, in the CLI path the sample calls. Do not put a modal in front of the first pixel.

When the inbox clusters, change the sample and note it in the changelog with the error code. Evaluators who hit the old failure can search the code and see that it moved.

That changelog line is worth more than a nurture email to people who never created an account.