SPB Git

Repository Public License

RELEASE-REPORT.md 12.0 KB

# ORIX Clean 1.0.0 (1) — Release Report

Date: 2026-11-21. Build machine: MacBook Pro Mac17,6 (Apple M5 Max, 18 cores, 48 GB), macOS 27.0 (26A5425a), Xcode 27.2, Swift 6.4.2.

Everything below was observed on that machine. What was not observed is listed under "Not verified".

# 1. Summary

Item Value
Version / build 1.0.0 / 1
Bundle identifier ai.spboucher.orix.clean
Minimum macOS 15.0
Architectures arm64, x86_64 (universal)
Repository Public since 2026-11-21, at the owner's request: https://git.spboucher.ai/orix-clean
Commit see git log -1 (the report is part of the release commit)
Developer ID signing Valid — Developer ID Application: Simon-Pierre Boucher (3YM54G49SN), Hardened Runtime (flags=0x10000(runtime)), secure timestamp, empty entitlements, no get-task-allow
Notarization — app Accepted — submission c2e7cbaa-9410-466f-b912-b9666f5a02b7, 0 issues in Apple's log, stapled and validated
Notarization — DMG Accepted — submission ff776146-c989-45d6-99de-c93bb217d8c1, stapled and validated
Gatekeeper accepted, source=Notarized Developer ID for the app, the DMG, and the app extracted from the ZIP
DMG ORIXClean-1.0.0.dmg, 5,133,225 bytes, SHA-256 c7a8b13f3db143e80d2902882d8fe069d68b6e89a1de212411dffdb7334bcd34
ZIP ORIXClean-1.0.0.zip, 4,032,161 bytes, SHA-256 126a980dd974136fc958ff7ab9a9b735d68aa2cc162d2729fa4f7d55a03da9fd
SPB Git release URL https://git.spboucher.ai/orix-clean/releases — assets under https://git.spboucher.ai/releases/orix-clean/v1.0.0/
Release tag v1.0.0 = aad1fa5 (the commit the binaries were built from; later commits change documentation only)

# 2. Implemented features

Feature State Notes
Storage analysis engine Implemented, tested getattrlistbulk, logical / allocated / private sizes, hard links once, clones and sparse files flagged, boundaries, cancel, pause
Interactive storage map Implemented Squarified treemap in Canvas, drill-down, breadcrumb, pinch zoom, drag pan, hover, search, context actions, list equivalent. Sunburst not implemented
Large-file discovery Implemented Files and folders; size, age, type, name, package filters
Duplicate detection Implemented, tested Size → sampled hash → SHA-256 → byte verification before removal
Similar images Implemented, tested on generated pictures Difference hash + Vision feature prints, fast and thorough modes
Cache cleanup Implemented, tested 12 documented rules, exclusions for system and stateful caches
Developer cleanup Implemented, tested 23 documented rules + build folders next to their manifest; simulators, runtimes, Docker reported only
Downloads cleanup Implemented, tested Installers, disk images, archives, incomplete downloads
Trash analysis Implemented, tested Home and per-volume Trash; emptying with typed confirmation
Uninstall assistant Implemented, tested Signature facts, associated files by certainty
Leftover detection Implemented, tested Name-based inference, labelled as such
Cleanup preview Implemented, tested (UI) The only entry to removal
Safe deletion engine Implemented, tested Authorization, safety policy, revalidation, five outcomes, Put Back
Storage intelligence Implemented, tested Evidence-based rules
Storage history Implemented, tested Samples while the app runs; watched folders; cleanup log
ORIX Clean Actions Implemented, tested Four built-in workflows, editor, review rule enforced
Reports Implemented, tested Nine report types; PDF, CSV, JSON, Markdown
Localization Implemented English, Canadian French (684 strings); no plural variants
Automatic updates Not implemented Evaluated (Sparkle) and deferred: the product has no network code
Pause/resume of duplicate and image analyses Not implemented Cancel only
Permanent deletion outside the Trash Deliberately absent

# 3. Verified functionality

  • Core tests: 56 tests in 10 suites, all passing (swift test, about 8 s). Core line coverage 86.7 % (2,662 / 3,070).
  • UI tests: 8 XCUITest scenarios, all passing, on a sample home folder stored on a disposable disk image. One scenario performs a real cleanup (preview → cancel → confirm → results → Put Back) and checks the filesystem at each step. See section 10 for the caveat on reruns.
  • Release build smoke test: the signed Release application was launched on a sample folder, scanned it and rendered the map; no crash report was produced. The debug-only screenshot hooks are absent from the Release binary (checked with strings).
  • Repository hygiene: Scripts/check.sh passes — author headers, no signing material, no credential pattern, no networking API and no privilege-escalation API in the product sources.

# 4. Storage accuracy

  • Allocated total of a 31,259-object tree (~/Library/Developer/Xcode) equals du -sk exactly: 2,412,920,832 bytes.
  • A test compares scanner totals with stat sums and with du on every run.
  • Hard-linked files are counted once (test); APFS clones report private sizes under 100 KB for a 4 MB clone (test); a 200 MB sparse file reports under 1 MB allocated (test).

# 5. APFS limitations

  • Private sizes are reported per file by APFS; a folder's private size is the sum of its files' and is therefore a lower bound of what removing the whole folder frees.
  • Local snapshots are not enumerated. Space held by a snapshot is not returned until the snapshot expires.
  • Purgeable space is shown as macOS reports it and is not attributed to files.
  • Moving to the Trash frees nothing until the Trash is emptied. On a full test volume, moving to the Trash still succeeds (it is a rename) — tested.
  • On filesystems that do not report sharing (the fallback path), recoverable space equals allocated size and the report says so. This path was exercised only through the API fallback, not on a real non-APFS volume.

# 6. Cleanup safety validation

Property Evidence
Nothing is removed without authorization nothingIsRemovedWithoutAuthorization; UI scenario checks the file still exists after selection and after cancelling the preview
High-risk items need a separate acknowledgement highRiskNeedsExplicitAcknowledgement
Protected paths are refused whatever asks protectedLocationsAreDenied (39 paths, including volume Trash folders, temporary roots and credential dot-folders), engine test on a substitute home
Items changed since the scan are left alone revalidationSkipsVanishedReplacedAndChangedItems
A duplicate is removed only if byte-identical to the kept copy duplicatesAreVerifiedAgainstTheKeptCopy
A duplicate group is never emptied aGroupIsNeverEmptied
No silent permanent deletion partialFailureIsReportedPerItem (no-Trash volume → manual action), permanent mode refused outside the Trash category
Restore never overwrites restoreRefusesToOverwrite
Workflows cannot remove without review validationForbidsRemovalWithoutReview, duplicateWorkflowKeepsOneCopyAndStopsWhenReviewIsDeclined, UI scenario
Read-only and full volumes readOnlyVolumeFailsCleanlyAndFullVolumeStillTrashes

No destructive test touches real user data: temporary directories and test disk images only. One core test moves a temporary folder to the real user Trash and restores it immediately.

# 7. Performance benchmarks

Release build of orix-bench, warm file cache, internal APFS SSD.

Measurement Result
Scan ~/Library/Developer (985,558 files, 699,525 folders, 8 threads) 17.4 s, 96,757 objects/s, 105 CPU-s, 202 MB peak RSS
Scan /Applications (448,727 files, 64,278 folders) — 1 / 2 / 3 / 4 / 6 / 8 threads 3.45 / 3.49 / 3.04 / 2.92 / 2.96 / 4.07 s; 3.4 / 6.8 / 8.9 / 11.2 / 16.7 / 23.4 CPU-s
Scan /System/Library (296,089 files, 159,923 folders) 3.27 s; 68 unreadable folders reported
Treemap layout (467,907 nodes, 1600×1000) 0.162 ms, 1,798 cells
Duplicate search on the sample folder 685 MB hashed in 0.069 s (warm cache)
Similar images on the sample folder 31 images in 0.21 s
Signed application bundle 13 MB

The default thread count (3; 1 in Low Power Mode or under thermal pressure) was chosen from the /Applications series. Application overhead while idle, cold-cache hashing throughput, cleanup latency on large selections and battery impact were not measured.

# 8. Supported Mac configurations

Configuration Status
macOS 27.0, Apple Silicon (M5 Max) Built, tested, run
macOS 15 and 26 Deployment target allows them; never run
Intel Macs x86_64 slice is in the binary; never run
Non-APFS, network and large external volumes Small APFS disk images only; otherwise not tested

# 9. Publication and approvals

  • The brief asked for a private repository by default. The source was first pushed to a private SSH-only location; on 2026-11-21 the owner asked for a public repository and a release, and both were done the same day.
  • Release assets: ORIXClean-1.0.0.dmg, ORIXClean-1.0.0.zip, SHA256SUMS.txt, each with a .sha256 sidecar.
  • Verification after upload: the three files were downloaded from the public URLs; sizes and SHA-256 match dist/, stapler validate succeeds on the downloaded DMG and spctl answers accepted, source=Notarized Developer ID.
  • README on the forge: the served page contains the 12 Mermaid blocks and every one of the 20 local images is served with HTTP 200. It was not inspected visually in a browser.
  • Apple Developer web pages (developer.apple.com, App Store Connect) were not opened: signing and notarization used the Developer ID identity and the notarytool keychain profile already installed. No agreement was accepted and no credential was created or changed.
  • Nothing is outstanding. The earlier private copy (gitsrv:srv/git-private/orix-clean.git) still exists and can be deleted.

# 10. Not verified

  • README rendering in a real browser (checked from the served HTML only).
  • Installation on a second, clean Mac. Gatekeeper was assessed with spctl on the build Mac only.
  • VoiceOver with a real user, Voice Control, measured contrast ratios.
  • French screens other than the storage map (strings are complete and checked for placeholders; layout was viewed on one screen).
  • Full-disk scan of this Mac in the application (the engine was run on folders up to 1.7 million objects).
  • Full Disk Access granted to ORIX Clean itself: the application was run without it; the protected-folder paths were exercised through permission-denied tests only.
  • UI test stability: the 8 scenarios passed, but during development several runs failed or hung because another application's test window took focus on the same Mac. The last full run is recorded below.

Final UI runs on the released code: a full run passed 7 of 8 scenarios (the eighth failed with "Failed to activate application" while another application's UI tests were running on the same Mac); that scenario then passed when run alone. The previous full run, one commit earlier, passed 8 of 8 in a single run. The final signed and notarized binary was launched once from dist/ and stayed up for five seconds without a crash report; the full scan smoke test was done on the Release build made before the last fixes.

# 11. Known limitations

See README, section 33. The main ones: one tested configuration; memory proportional to the number of objects; scan results not persisted; recoverable space is an estimate; associations and leftovers are inferences; root-owned items cannot be removed; history only while the app runs; similar-image thresholds calibrated on generated pictures; no plural forms in strings; no automatic updates; not sandboxed.

# 12. Remaining work

  • Run on macOS 15, macOS 26 and an Intel Mac.
  • Try the app on a real home folder with Full Disk Access and on a real photo collection.
  • Plural-aware strings; background sampling; layered icon for macOS 26 and later.
Files main
Menu

Theme