[dmoliveira]

Summary

• Retain profile case roots and version-probe roots on success and failure; report retained basenames, case teardown and fixture-identity verification separately.
• Preserve the existing cleanup/qualification gate while making cleanup_complete mean verified teardown and fixture identity, not deletion. Failure paths remain unqualified.
• Document that this older probe still writes witnesses, traces, configuration and tripwires by pathname: independently verified OS isolation is needed when preserving local data matters. The separate older dependency-boundary probe still deletes roots and is not covered by this change.
• Keep the probe limited to OpenCode 2.2.5 and leave all gateway and dependency authority disabled.

Validation

• 87 focused profile, auxiliary-redaction and direct-shell tests; Ruff; Python compilation; git diff --check; static pre-commit hooks.
• A full 18-case tmux-backed run on the selected OpenCode 2.2.5 executable (SHA-256 abe4d88594bffb12b364375b8a6732d2d1f7a977195ed1a35c5c76f8e89c9e0e) returned measurement PASS and profile UNQUALIFIED. All 18 case roots and the version root were retained; checked process/socket teardown completed.
• The macOS sandbox test allowed fixture writes and loopback while denying outside and symlink-target writes, non-loopback connections and external signals. This is bounded evidence, not complete containment or gateway protection.
• Local pre-commit make validate was skipped because it shares a fixed-name configuration lock and executes older recursive-cleanup tests; exact-head CI supplies full validation and self-tests in an isolated runner.