> ## Documentation Index > Fetch the complete documentation index at: https://aidocs.ethanbragdon.icu/llms.txt > Use this file to discover all available pages before exploring further. # FORGE, Research & Knowledge Troubleshooting > Troubleshoot Starfire AI builds, artifacts, research runs, Knowledge indexing, retrieval, and long-running job failures. # FORGE, research & Knowledge troubleshooting These systems share a common characteristic: they can involve background jobs, multiple model/tool calls, indexing, or long-running state. A successful click or accepted request does not always mean the final job completed successfully. ## FORGE build is stuck Check the build state first. Possible causes include: * queued but no worker has claimed it * waiting on a model/provider * validation still running * repair loop active * packaging/upload in progress * exhausted build budget or limit * worker or queue degradation If multiple builds are stuck in the same stage, inspect Control Center health rather than retrying each user job manually. ## Build completed but no artifact appears Determine whether the build reached packaging successfully. A job can complete generation but fail while packaging or uploading the artifact. Check artifact metadata for failed upload, missing storage object, or expired/cleaned output. ## Validation keeps failing Review the actual validator output and repair-attempt history. A project can be generated correctly at a text level but still fail because of dependencies, configuration, unsupported environment assumptions, or an incompatible requirement. Do not report the build as validated if the validator never succeeded. ## Research run is slow or incomplete A deeper research run can involve planning, multiple searches, source retrieval, verification, and report generation. Check whether the delay is in: * search-provider access * source retrieval * model execution * verification * report generation * a configured source/depth limit ## Citations do not support the claim Treat that as a quality issue even if the source exists. Citation verification should compare the actual claim with the cited passage rather than assuming retrieval equals support. ## Knowledge source will not index Check: * source/file processing status * extraction failure * unsupported content * embedding/index provider health * storage state * organization/project permission * stale or failed indexing record Authorized operators can retry indexing or rebuild supported indexes from Knowledge administration. ## Retrieval misses a known document Possible causes: * document not indexed * query does not match the relevant chunk * reranking removed the candidate * source filtered by project/organization scope * permission denial * context/result limit Use retrieval inspection tools where available instead of repeatedly re-uploading the same source. ## Escalate Escalate when failures affect many users, many jobs are stuck at the same stage, storage/index workers are unhealthy, or provider-level errors are widespread. Use Control Center to inspect build, research, artifact, Knowledge, retrieval, and tool health.