When something is wrong
Every failure below is one that actually happened, with what it looked like at the time. StellarView tries hard not to report success it cannot verify, so most problems announce themselves. These are the ones that do not.
The first screen says a key is missing and I set it
Check which file you set it in. Keys go in config/.env in the directory you ran docker compose up from. That directory is mounted over /app/config inside the container, so anything shipped in the image at that path is invisible.
If you set it in the app instead, it takes effect immediately, with no restart. Seeing the old state usually means a second container is still running:
$ docker compose psThe glossary is empty, or the scroller says it cannot load
/api/glossary reads docs/glossary.yml. If it returns 500, that file is not in the image. An early build shipped without it, and the first screen could not define its own words. Pull a current image.
There is no fallback vocabulary on purpose: a stale copy of the language is worse than none, and that is how one term came to mean two different things.
"Just show me" is disabled
The door reports its own reason. If no demo repository is configured it names the setting — DEMO_GALAXY_REPO — rather than doing nothing when clicked.
Generate Spec is greyed out
Client Domain is required and empty. The button stays disabled until it is filled. That is deliberate: a plan built on a blank required field is worse than no plan, because it looks finished.
The JIRA test passes but nothing reaches JIRA
If the test really passed, it read your account and your project. If you never ran it, run it now. Some Atlassian endpoints answer an unauthenticated request with an empty success rather than a 401, so a connection can look fine while the token is not being applied at all.
Most common cause: Atlassian wants a classic API token and your account email, not a username and not a scoped token.
A clone succeeded but the code is not what I expected
Check the branch. A clone that lands on the default branch looks perfectly healthy and can contain none of the work. That failure shipped here once and reported success every time. GITHUB_SOURCE_BRANCH in the galaxy's .env sets which branch it should be on.
It says healthy and the page is blank
A health check that cannot see the thing you care about will still return green. If the service is up and the interface is not there, the assets did not make it into the image. Check the browser console before trusting any status indicator, a load balancer and a browser disagree more often than you would like.
Analysis finished but the counts look wrong
Re-run it. Analysis is a snapshot, and a large merge since import will not be reflected until you read the repository again.
Something else
Two places tell you more than a status light:
$ docker compose logs -f the run, as it happensand the browser console, which is the only thing that sees the interface actually render.
