Getting started
2 min read
Read a release brief
Understand source risk, delivery, distribution, and runtime health without mixing them together.
Before you start
- An app with at least one release record.
Step by step
Choose the platform and version
Use the app, platform, and version selectors. The version’s identity determines which evidence can be attached.
Check what is known
Delivery describes a recorded pipeline result. Source risk describes analyzed code changes. Store context describes distribution. Runtime health requires runtime evidence; delivery success alone is not enough.
Review findings and their evidence
Read the finding, impact, and recommended next check. A confidence score is not a guarantee. A source comparison requires a truthful baseline and target commit.
Investigate with your coding agent
Expand a finding and choose Copy fix prompt. Paste it into an agent connected to Specita through MCP, with the matching repository open. The agent reads the exact finding and its evidence, checks the code, and uses its separate repository access to work on a justified fix.
Record the team’s response
Assign a response owner, acknowledge the issue, and add a resolution note before resolving. This records a human decision without rewriting the underlying evidence.
Check your setup
- The release belongs to the expected app and platform.
- The comparison baseline is appropriate for the environment.
- Unknown or stale evidence is not treated as healthy.
Troubleshooting
No source comparison
Attach a repository and valid commit identity. Use the deployment timeline to select the previous successful version in that environment.
Brief is partial
Read the visible stage or error. Missing evidence, provider failures, and package limits need different recovery steps; do not assume the deployment failed.
Version share unavailable
Specita does not yet calculate a version’s share of recorded sessions. The crash-free percentage describes recorded sessions without a crash, not the percentage of users on a version. Store rollout settings do not measure actual use.
Credential safety
Provider credentials are not part of the model’s analysis context. Keep secrets out of source files and release notes.