Release readiness

This is a verification record and release checklist, not a claim that every supported target has been tested on a user's machine. Last reviewed: 2026-09-04.

Version scope and release verification

Category filtering, cumulative filtered selection, and Shift+A global selection apply to 0.15.0 and later. The compatibility notes below describe the restore-record changes in 0.15.0. See the changelog for details.

Consult the v0.15.0 release for publication status and its per-platform release-evidence.json. The table below records historical source checks and remaining Trash acceptance; it does not replace installation checks on 0.15.0 assets. The recorded walkthrough retains its v0.14.0 version label. A passing check on a later commit does not retroactively validate an earlier release asset.

Support and verification matrix

Platform	Native distribution targets	Evidence available	Remaining acceptance
macOS	Apple Silicon, Intel	The source test macos_system_trash_round_trip_records_exact_locator passes in the macOS job of CI for e136149. It moves and restores a test-owned directory with Chinese text and spaces, checking the exact Trash locator and nested file contents.	Repeat using the actual downloaded release package and supported OS/volume combinations. The source test is not proof for every filesystem or package.
Windows	x64, x86	Recycle Bin cleanup/restore code and source CI exist. v0.14.0's source CI had two Windows TUI assertion failures; subsequent fixes are not part of that release.	Real Recycle Bin round trip and restored-content checks on Windows, plus downloaded-package first-run validation.
Linux	x86-64, ARM64, ARMv7 hard-float	Freedesktop Trash cleanup/restore code and source CI exist.	Real desktop Trash round trip and downloaded-package validation. Headless systems, mount policy, and Trash availability need explicit testing.

Distribution targets are not a promise of identical cleanup coverage. The Windows routine system scope is deliberately narrow; macOS has more named application and developer-cache locations. A passed unit test, artifact creation, or successful version command does not establish full cleanup/restore behavior.

Updating and compatibility

Use the installation guide to keep the install method consistent. Preserve local configuration and history before an upgrade if you may need them. Do not assume an older binary can read a newer plan or manifest. Never hand-edit a saved plan to bypass a compatibility error: create and review a new plan with the version you intend to run.

Compatibility change in 0.15.0: new restore records use cleanr.restore.v2; the 0.15.0 reader also accepts v1 history. not-attempted means the backend was not called, while pending means the intent was saved but the result was not recorded. After an interruption, pending items require manual inspection of the original path, Trash, and manifest before retrying. Do not roll back to an older binary to process v2 records; retain the state directory and use the current or a newer compatible version.

In 0.15.0, cleanup and restore hold an OS file lock in the state directory for the operation. Another operation reports a conflict; the OS releases the lock when the process exits, so do not delete the lock file to bypass it. Trash and JSON writes still cannot be one atomic transaction. If recording a filesystem result fails, later items stop and the reported run/manifest details support manual recovery.

Binary rollback does not undo file moves, restore history, or configuration changes. Keep system Trash and manifests until recovery is no longer needed.

Before publishing

Use the repository's release procedure. Record the exact commit, package version, platform, and evidence for each gate:

• Required source checks and package smoke checks pass for the release commit.
• Each promoted installation path can run --version, --help, and an isolated read-only analysis. Test upgrade/removal separately; retain recovery records.
• The selected platform's real Trash/restore scenario passes with generated samples, including original-path conflicts and interrupted restoration.
• The deployed /Cleanr/ website, CSS/JS, English and Chinese installation pages, and shared-card image respond successfully. A local fix alone is not deployment.
• Release notes distinguish published features, pending features, format changes, and known limitations. Verify the public GitHub, npm, and crates.io version after an explicitly authorized release.

A real Trash test changes the system Trash and needs its own execution scope; read-only analysis or simulated backend tests do not authorize it.

Before promoting

Present Cleanr as AI-friendly disk cleanup under the user's control: application and browser caches, logs, temporary files, downloads, and development artifacts. Show how structured, read-only evidence helps an agent explain candidates and prepare a reviewable plan, alongside the direct terminal workflow. Explain safety through default confirmation, plan and target checks, system Trash, and local recovery records. For each walkthrough, choose one scope and show the candidate reasons, human selection, and a recoverable move to Trash. Use synthetic or redacted paths, show review and recovery as separate actions, and identify the app version and platform. Label repository-only visuals Unreleased.

Show candidate bytes and moved bytes accurately. Do not equate them with newly available disk space or promise guaranteed recovery. Present the TUI and agent entry points together, with the cloud-agent data boundary visible.

Start with a small audience comfortable with terminal tools, covering both everyday cleanup and development use cases. Track installation success, time to the first understood review, empty-result confusion, unexpected rule matches, and restore failures through voluntary feedback. Baselines and real-user demand are not yet measured here; local scenario success cannot replace that evidence. Use support channels and private security reports. Do not request raw analysis JSON. Posting announcements or contacting users is a separate publication decision.