Separate successful live QA from release dependency parity

A full manual integration run can pass every exposed action while the host preview installs a different SDK version from the repository pin. Record the actual startup dependency version as live evidence and treat package parity as a separate release check. For revision restore, comparing provider content hashes before overwrite and after restore gives stronger evidence than a green success card; cancellation should be matched to absence of a tool request, not only a narration message.