Troubleshooting
Start with the first failed checkpoint. Record the time, build, workspace and safe error message. Keep the saved draft and installation data while diagnosing.
Console does not open
Section titled “Console does not open”Confirm the operator-supplied URL and whether its hostname resolves from your
machine. Ask the operator to inspect the Console and HTTPS edge, then the
configured health endpoints. A port on another machine is not your localhost.
Do not use a public Docker image’s internal port as a substitute for the installation’s configured HTTPS origin.
Certificate warning or TLS connection error
Section titled “Certificate warning or TLS connection error”Check hostname, expiry and trust for the actual connection. Browser-to-Console, Console-to-identity/API and API-to-runner connections may have different trust configuration. A certificate trusted by your browser is not automatically trusted inside a container.
Have the operator install or configure the correct public trust material. Do not disable certificate verification. Revision-signing trust is separate from HTTPS trust.
Sign-in loops, 401 or 403
Section titled “Sign-in loops, 401 or 403”- A login loop can indicate mismatched external origins, redirect URLs, certificate trust, session keys or identity-provider configuration.
401means the request did not establish acceptable authentication.403means access was denied; check exact tenant/workspace claims and the required role or policy with the operator.
Anonymous rejection of protected API routes is expected. Have the operator correct the account’s access for the intended workspace, then start a new login. Keep authentication enabled and resource permissions limited to the task.
Migration command fails
Section titled “Migration command fails”Keep the command’s safe error and the selected image digest. Verify the intended database, credentials and the exact reviewed plan hash. Resolve the cause before rerunning. Do not substitute a plan from another image or delete the database to clear an unexplained failure.
Test draft reports the runner unavailable
Section titled “Test draft reports the runner unavailable”Check that the development draft runner, not only Enterprise Node, is running.
The API needs a reachable HTTPS address, valid runner credential, trusted
certificate and matching exact workspace scope. A container’s localhost does
not reach a runner on the Windows host.
After the operator corrects the connection, choose Refresh, then run the synthetic source. For browser execution, check the interactive Windows desktop and WebView2 availability too.
The test fails before my code runs
Section titled “The test fails before my code runs”Verify the source descriptor’s module path and export, input and output schemas, execution profile and test input. Use the example files as a known small contract. A setup error is different from an exception thrown inside the extractor.
The code runs but the result is wrong
Section titled “The code runs but the result is wrong”Confirm which saved version and input ran. Inspect the returned records rather than relying on the editor text or a previously selected sample. Check field names, types and the output envelope. Follow source locations when shown, save the correction and test again.
Save conflict, long test or missing result
Section titled “Save conflict, long test or missing result”- Save conflict: review the latest saved version before reconciling edits.
- Long-running test: choose Stop test, wait for its terminal outcome and inspect the code before starting again.
- Missing result after reload/restart: saved code and temporary run output have different lifetimes. Confirm the previous outcome before replaying work.
- Preview has data but the product API does not: draft samples are previews. Check release, activation and accepted-record state with the operator.
AI does not generate code
Section titled “AI does not generate code”Check whether inference is enabled for your workspace and whether the selected route and model match the operator’s configuration. Saving settings does not contact the provider; Test connection makes a separate inference request and can consume tokens. Provider integration is experimental: see AI and coding assistants for its current status. You can continue editing and testing drafts manually while the connection is investigated.
Report a problem
Section titled “Report a problem”Send your agreed support contact: the product and build, the screen/action, reproduction steps, expected and actual result, time and a safe error excerpt. Use synthetic inputs. Do not send access tokens, signing keys, cookies or confidential records. For public example or documentation problems, use the feedback repository. Keep installation-specific material in the private support channel agreed with your operator.