SWITCH CLEAN 0.7.0 - DISPOSABLE WINDOWS VM VALIDATION
November 12, 2026

PASS: 93 regression tests on the final build, both on host disposable fixtures and
inside the SAC-enabled Windows VM. No host cleanup or driver mutation was run.
PASS: Native portable app launched with SAC enabled, both elevated and standard
user, without installation, Run anyway, exclusions or disabling SAC.
PASS: All 37 bundled EXE/DLL/PYD files retain valid publisher signatures.

LIVE VM ACTIONS (phase 9)
- Original 91 tests passed before live actions.
- Old browser-cache and crash-dump fixtures deleted; cookies/login fixture data,
  offline site data, recent dump and an autosave document remained byte-identical.
- Windows Temporary Setup Files and Thumbnail Cache handlers returned success.
  Per-handler deleted bytes are unknown; Windows may find no eligible content.
- Windows media-appropriate fixed-volume optimization completed successfully,
  including its reported retrim. This is not a claim about physical SSD behavior.
- Full recovery exported all 3 guest OEM packages; focused archive exported exactly
  the explicitly selected Realtek fixture (1 package). The selection tests subset
  export, not a claim that the VM needs that driver for its own hardware.
- Both archives loaded using the production verifier; all 16 exported file hashes
  were independently rechecked from the stopped guest disk.
- Both protected package removal requests were rejected before mutation.
- Guest device/driver fingerprint unchanged.
- CiTool showed VerifiedAndReputableDesktop enforced before and after live work.
  The user approved one normal PSF UAC prompt in QEMU; no SAC bypass was used.

FINAL-BUILD FOLLOW-UP (phase 10)
Visual QA discovered a preliminary Print to PDF label still said removal candidate,
although the final guard blocked deletion. The final build publishes the protection
in reinstall assessment/removal_eligible/focused_reason as well. Protection logic
and the cleanup/export execution adapters are unchanged by this correction.
- All 93 tests passed in a standard-user process, without another UAC prompt.
- Live printer review displayed Protected - keep and removal_eligible=False.
- A protected removal request was rejected; driver inventory unchanged.
- Existing full and focused archives loaded successfully with the final code.
- Registry SAC state was Enforced before and after. CiTool policy enumeration was
  denied to this standard-user process; authoritative policy enumeration is from
  the preceding elevated live run in the same unchanged VM.
- Final app/test Python source hashes match the tested files read independently
  from the stopped VM disk. Native UI was inspected in light mode in the VM and
  dark mode on the host, with both top-level workflows and their primary buttons.

HOST PROTECTION
All 148 installed OEM INF names and SHA-256 hashes match the pre-test inventory.
No host driver/device removal, restore, cleanup, Razer/THX change or security change.
The existing 71-package local focused profile is preserved, outside the portable ZIP.
The desktop shortcut is the only user-facing host configuration updated.

LIMITS
No real Windows.old rollback contents, active upgrade, vendor app version matrix,
physical SSD discard effectiveness, or production recovery restore was exercised
in this run. Recycle Bin and restore live-workflow evidence from prior versions is
retained in VM-VALIDATION-HISTORY.txt; this release reran their isolated tests.
Windows-managed cleanup totals are not independently measured. The final build's
cleanup execution code matches the elevated live test; the follow-up corrected
review metadata and added two regression tests. No claim is made that Microsoft's
future reputation/policy decisions or every downloaded distribution will behave
identically. No installer, unsigned replacement host or new third-party runtime
was introduced. VM sign-in was automatic; the VM was normally shut down afterward.

EVIDENCE
validation/vm-results.json: live operations, policy enumeration, backup manifests.
validation/final-build.json: final 93 tests, corrected review, source hashes.
validation/independent-verification.json: stopped-disk and host checks.


SWITCH CLEAN 0.6.0 â€” DISPOSABLE WINDOWS VM VALIDATION
2026-11-12

PASS: 76 isolated regression tests executed inside the Windows VM.
PASS: Smart App Control VerifiedAndReputableDesktop policy enforced before and after.
No Run anyway, SAC disablement, application-control exclusion, or policy bypass.
A normal Python Software Foundation UAC prompt was approved by the user.

Live guest checks:
- Existing printer/spooler package protected.
- Both protected package selections rejected by the public engine before mutation.
- Full recovery exported all 3 installed OEM driver packages.
- Focused backup exported exactly 1 explicitly chosen Realtek test fixture.
  This selection exercises subset backup; it is not a claim that the VM needs that driver.
- Both archives loaded successfully using the production restore verifier.
- All 16 file hashes independently verified from the stopped VM disk.
- Guest driver/device inventory unchanged by the tests.

Isolated fixtures executed inside the SAC-enabled VM additionally covered:
Print to PDF without PnP associations; VPN/demand-created adapters; present hardware
using a different INF; internal hardware alternatives; shared/unknown services;
extended device associations; incomplete inventory; incoming Include dependencies;
protected extension classes; reviewed focused-selection profiles; protected UI
selection; and repeat checks before removal, backup failures and Windows refusals.
These are fixture tests, not tests with every vendor's physical hardware or VPN app.

VM automatic sign-in repaired using its existing account and password, with the
login-count limit removed. Automatic arrival at the guest desktop was observed.
Only disposable VM disk copies and task evidence files were modified.
Host settings, applications, shortcuts, drivers and security policies were not changed.
All 148 host OEM INF hashes still match the pre-test baseline.
The VM was shut down normally after testing.

Evidence:
SWITCH-Clean-0.6.0-VM-results.json
SWITCH-Clean-0.6.0-VM-verification.json



HISTORICAL VALIDATION - prior versions only



SWITCH CLEAN 0.2.0 - DISPOSABLE WINDOWS VM VALIDATION


2026-11-11 | Windows 11 Home x64, image version 10.0.26300.9711





RESULT: All five exercised operations passed under Smart App Control enforcement.





Smart App Control


- Live registry status: Enforced before and after the run.


- CiTool reported VerifiedAndReputableDesktop with IsEnforced=true.


- Evaluation policy was not enforced. No policy was disabled or bypassed.


- User enabled SAC and approved the normal administrator prompt.


- The test host retained the original Python Software Foundation signature.





PASS: User and system temporary-file cleanup


- Deleted 132,820,892 bytes in total (about 126.7 MiB).


- Both 10-day-old test files were removed; both recent files remained.


- The real engine scanned and deleted eligible guest temporary files.





PASS: Selective driver backup


- Guest contained two published OEM packages; selected only oem0.inf.


- PnPUtil exported the Microsoft printer package prnms009.inf.


- Only the selected package directory was created; manifest.json was present.


- All seven exported files matched their SHA-256 hashes, checked both by the


  guest runner and independently by reading the VM disk from the host.





PASS: Non-present device removal


- All four scanned non-present instances were removed successfully.


- A fresh inventory found none of the selected instances remaining.


- All 58 previously present device instances remained in the inventory.


- Live removal coverage was the Processor class (disconnected emulated CPUs).


  Enumeration is across all classes; this is not live coverage of every class.





PASS: Component-store cleanup


- Actual DISM /Online /Cleanup-Image /StartComponentCleanup /NoRestart ran.


- Exit code 0; DISM reported successful completion and no restart required.


- No ResetBase operation was used.





PASS: Windows-volume retrim


- Actual defrag.exe C: /L /U /V ran and returned exit code 0.


- Windows reported successful completion, 9,730 allocations trimmed and


  67.20 GB of trim ranges. This is not 67.20 GB of newly freed space and does


  not demonstrate physical SSD behavior: the disk is virtual.





Scope and remaining gaps


- This was an elevated integration test of the shipped core.py, winapi.py and


  safe_temp.py through a separate guest-only harness. It complements earlier


  native UI launch/scan tests; it is not end-to-end GUI coverage of every action.


- Recycle Bin confirmation/emptying and Windows Disk Cleanup dialogs, including


  old Windows installations, were not exercised in this run.


- Driver restore, dependency completeness, non-Microsoft vendor packages,


  post-reboot device behavior and other device classes remain unqualified.


- No driver-store pruning is implemented or claimed.


- Test runner binaries and application files on the VM test ISO were local test


  artifacts; the earlier download-zone-marked launch test is a separate result.





Isolation and evidence


- A new QCOW2 overlay held all Windows changes; backing images were read-only.


- No physical host drives or host folders were shared with the VM.


- The host's cleanup routines were not invoked. Guest maintenance only.


- Raw results: SWITCH-Clean-VM-evidence/results.json


- Independent checks: SWITCH-Clean-VM-evidence/backup-hash-verification.json


- Tested engine hashes: SWITCH-Clean-VM-evidence/tested-source-hashes.json


- A normal guest shutdown was requested after evidence collection.





These results qualify the listed operations on this VM, not all hardware,


Windows builds, device classes, enterprise policies or cleanup scenarios.





PHASE 2 â€” 2026-11-11, SAC Enforced before/after


Recycle Bin: cancel preserved the 8,800-byte fixture; confirmation emptied it.


Windows Disk Cleanup: cancel preserved the fixture; selecting Recycle Bin and


confirming Delete Files removed it. Driver-package category was unchecked.


Four hotplug fixture instances removed across DiskDrive, HIDClass, Mouse and USB;


no selected instances remained, no previously present instances disappeared,


and the two OEM packages were unchanged. Live operations used 0.2.0 engine.


FOUND: SHEmptyRecycleBin returned S_OK on cancellation. The old report incorrectly


said success. 0.3.0 now checks the remaining count and reports unsuccessful when


items remain or verification is unavailable. Live regression passed in Phase 3.


See Phase 3 for driver-package validation limits.





PHASE 3 â€” 2026-11-11, 0.3.0 engine


PASS: Recycle cancellation retained the fixture and correctly reported unsuccessful;


confirmation subsequently emptied the bin and reported success.


PASS: Driver package oem0.inf (Microsoft prnms009.inf) exported seven files; INF


matched the preview, hashes were verified, and the recovery manifest was saved.


EXPECTED SAFETY BEHAVIOR: Windows rejected non-forced removal with 0xE000023D,


reporting installed devices using that INF. The application reported failure,


retained the backup and did not retry with force/uninstall. Both OEM packages


remained. Independent read-only inspection of the shut-down VM verified all seven


backup file hashes and both retained OEM INFs.


TEST FIXTURE LIMIT: The harness expected a deletable package and stopped on an


assertion after the refusal. This is not a successful live deletion test. No


suitable unused third-party package was available in this VM. Actual successful


Driver Store deletion, vendor dependencies and recovery remain unqualified;


automated tests cover successful deletion, refusal, stale associations, backup


failure and mismatches. This was the Phase 3 limitation; successful live deletion was subsequently


validated in Phase 4 below. Vendor dependency and recovery coverage is still limited.


SAC: Live CiTool VerifiedAndReputableDesktop policy was enforced before Phase 3.


After normal VM shutdown the persisted active control set still had


VerifiedAndReputablePolicyState=1. No security settings were changed.


Evidence: phase3-results.json and phase3-independent-verification.json in the


sibling SWITCH-Clean-VM-evidence folder. No host maintenance was performed.





PHASE 4 â€” 2026-11-11, SUCCESSFUL UNUSED DRIVER-PACKAGE DELETION


Application: unchanged SWITCH Clean 0.3.0 engine, disposable Windows VM only.


Fixture: Realtek USB Ethernet, original INF rtucx22x64.inf, version


11.19.0602.2025. Its catalog has a Valid Microsoft Windows Hardware Compatibility


Publisher signature. A copy was exported read-only from the host; host drivers


were not installed, removed or altered. Fixture is test media, not app payload.





Before staging, PnPUtil /enum-devices /deviceids was checked against all USB


VID/PID IDs from the fixture INF; none matched any enumerated device. The fixture


was staged with /add-driver only (no /install), published as oem2.inf. Device IDs


were unchanged, the package had no present or absent device associations, and


SWITCH Clean's normal eligibility checks accepted it.





PASS: Engine exported the reviewed package, matched the INF to the scan, verified


three backup file hashes, saved the manifest, rechecked eligibility, and called


PnPUtil /delete-driver oem2.inf without force, uninstall or reboot flags.


PASS: Windows returned exit 0, 'Driver package deleted successfully.'


PASS: oem2.inf disappeared; original oem0.inf and oem1.inf remained; device-instance


IDs were unchanged. No reboot was required by the operation.


PASS: Live CiTool VerifiedAndReputableDesktop IsEnforced=true before AND after;


registry SAC state Enforced. No security policy/settings changes were made.


PASS: After normal shutdown, independent read-only disk inspection verified all


three exported files against both the original fixture and the backup manifest,


confirmed selected INF absence and original INF retention. Persisted active


control-set VerifiedAndReputablePolicyState remained 1.





Evidence: sibling SWITCH-Clean-VM-evidence/phase4-results.json and


phase4-independent-verification.json. Test harness passed all assertions.


This validates one unused vendor package and the previous in-use refusal case;


it does not establish comprehensive vendor dependency detection or restore


compatibility. No working-host cleanup or driver changes were performed.





PHASE 5 â€” 2026-11-11, BACKUP RESTORE VERIFIED


Restored the Phase 4 SWITCH Clean backup, not the original fixture media, inside


the disposable VM. Checked the saved manifest and all three file hashes before


running PnPUtil /add-driver on the backed-up rtucx22x64.inf. No /install or force


flags were used. Windows accepted the package (exit 0), published as oem2.inf.





PASS: Restored published INF hash matched the backup. Re-exporting oem2.inf using


PnPUtil produced the CAT, INF and SYS files, all byte-identical by SHA-256 to the


backup. Existing oem0.inf/oem1.inf and all device-instance IDs were unchanged.


PASS: Live CiTool VerifiedAndReputableDesktop IsEnforced=true before and after;


SAC state Enforced. No reboot was required by restoration.


PASS: After normal shutdown, independent read-only VM disk inspection confirmed


the restored INF, all three re-exported hashes, original package retention, and


persisted VerifiedAndReputablePolicyState=1. Restored package retained in VM.





Scope: successful recovery to the Driver Store for this one signed Realtek USB


Ethernet package. The VM has no matching hardware, so device binding, loading and


network operation were not tested. This was a PnPUtil recovery test, not a restore


wizard added to SWITCH Clean. No host drivers or security settings were changed.


Evidence: sibling SWITCH-Clean-VM-evidence/phase5-results.json and


phase5-independent-verification.json.





PHASE 6 â€” 2026-11-11, NATIVE SELECTIVE RESTORE RELEASE 0.4.0


PASS: Actual 0.4.0 Engine.load_backup/execute restored one selected Realtek package


from a two-package SWITCH backup. Printer oem0.inf was left unselected. Observed


exactly one PnPUtil /add-driver <private selected INF> /install call, no wildcards,


force flags or automatic reboot. Windows returned success (0), no restart needed.


PASS: Re-exported CAT/INF/SYS hashes matched the selected backup. Existing device


IDs and original OEM packages remained unchanged. Corrupted backup CAT rejected.


PASS: CiTool VerifiedAndReputableDesktop enforced before AND after; SAC Enforced.


Independent disk inspection after normal shutdown verified all three re-exported


hashes, restored published INF, original package retention, and active persisted


VerifiedAndReputablePolicyState=1. No working-host driver changes were made.


PASS: 47 automated tests, including selective staging, corruption, unsafe paths,


changed previews, Windows refusal, expired plans and administrator checks.


PASS: Native Driver restore page and real verified-backup load visually inspected


on the host with SAC Enforced. Native theme/broom/portable no-console launch kept.


The live VM test invoked the production engine; native GUI loading was checked


separately. This was not a mouse-driven destructive GUI test.


LIMIT: No matching Realtek hardware exists in this VM. Device binding/loading and


network operation are not validated. Windows controls matching and driver ranking.


Evidence: sibling SWITCH-Clean-VM-evidence/phase6-results.json and


phase6-independent-verification.json. Runtime binaries are unchanged.





PHASE 7 â€” 2026-11-11, PREPARE FOR REINSTALL 0.5.0


PASS: Production assessment correctly classified the unused Realtek fixture as


not recommended. Test deliberately selected it as a user override to exercise


focused subset creation; it was not represented as needed by this VM.


PASS: Preparation exported all 3 published OEM packages into Full-recovery


(13 files), then copied only the checked package into Focused-reinstall (3 files).


Both manifests loaded successfully through the production Driver restore loader.


Focused files matched the full archive. assessment.json and reports were saved.


PASS: Device and OEM-package inventory fingerprints were unchanged before/after.


No drivers/devices were removed or installed by preparation.


PASS: Live CiTool VerifiedAndReputableDesktop IsEnforced=true before and after;


SAC Enforced. Independent read-only disk inspection after normal shutdown verified


all 16 archived file copies against their manifests. Active persisted


VerifiedAndReputablePolicyState remained 1.


PASS: 61 automated tests, including association classification, Include/SWC links,


exact extension matches, older alternatives, full/focused manifests, failed full


exports, changed inventories and empty focused selections.


PASS: Native preparation page and real read-only host scan visually inspected


with SAC enforced. SWITCH theme and broom retained; no installed dependencies or


own signing. All 37 shipped runtime binaries retain Valid Authenticode signatures.


Host snapshot: 161 OEM packages; initial 71 in-use + 3 supporting recommendations.


This is a time-specific inventory, not a fixed count or claim that all remaining


packages are unnecessary. The working host was scanned only, not cleaned/backed up.





LIMITS: VM test used a small three-package inventory. Hardware-specific dependency


completeness, live device operation, different Windows releases and a full OS


reinstallation were not tested. Recommended focus is conservative, not a guarantee


of the minimal/complete driver set. Uncertain packages require review. Full


recovery means OEM driver packages, not inbox drivers, applications or an OS image.


Evidence: sibling SWITCH-Clean-VM-evidence/phase7-results.json and


phase7-independent-verification.json. GUI preview tested separately from the


production-engine archive test in the VM.




SWITCH Clean 0.7.1 - disposable Windows VM validation
November 12, 2026

PASS: all 104 tests in the SAC-enabled Windows VM, including 11 new temporary-file
exclusion tests. The same 104 tests passed locally against disposable fixtures.
PASS: SAC registry state Enforced before and after; no policy/exclusion change or
Run anyway. Tests ran as a standard user through the unchanged signed portable host.
PASS: guest device/driver inventory unchanged.
PASS: 23 app/test Python source files independently read from the shut-down VM disk
match this release byte-for-byte. JSON result evidence is in validation/.

New tests cover preservation of complete current scanner refresh folders, pointer
changes after preview, missing/malformed/escaping pointers, missing or busy refresh
locks, lock release, a newly-running scanner, protected NVIDIA/Razer/THX paths,
unknown/nested cache layouts and rechecking exclusions on an existing preview.
Only recognized unreferenced engine-/refresh- staging can become a temp candidate.
Current packages, control files and scanner download archives remain protected.

The Windows cleanup/driver backup/restore adapters are unchanged from 0.7.0. Their
previous live-operation evidence and limitations are in VM-VALIDATION-HISTORY.txt.
This patch does not claim those live workflows were repeated. The new exclusion
logic was tested on disposable fixtures and then used for the separately authorized
host cleanup with protected-file/driver hash verification. Host audit reports are
private and excluded from the portable ZIP.

No new authoritative CiTool policy enumeration was attempted in this standard-user
VM run. The preceding elevated 0.7.0 run established VerifiedAndReputableDesktop
enforcement in this same disposable VM, and this run checked Enforced state at
both boundaries without changing security settings. Future Microsoft reputation
or policy decisions on other systems/distributions are not guaranteed.


SWITCH Clean 0.8.0 - disposable Windows VM validation
November 12, 2026

PASS: 135 regression tests locally and in the final standard-user VM run.
The elevated engine run passed the preceding 134 tests; the added 135th test
covers the final native results-screen transition fix.
PASS: SAC Enforced before and after. CiTool confirms VerifiedAndReputableDesktop
is enforced at both boundaries. No security policy, exclusion, download marking
or Run anyway bypass was used. Normal UAC was handled by the user.
PASS: Administrator scan measured all ten allowlisted Windows cleanup categories
using Windows COM handlers, and performed read-only component-store analysis.
PASS: Cleanup waited for the servicing worker's natural exit ([120.4] seconds).
No service was stopped. Cancellation, timeout, restart and unknown-state paths
were covered by isolated regression tests; no host restart was initiated.
PASS: Selected Windows Update, Delivery Optimization, setup-log and recovered-file
handlers ran together in one isolated profile. Temporary profile flags were removed.
PASS: Live aged-temp, setup-log and recovered-fragment fixtures were removed.
Brave, Razer, THX and NVIDIA fixtures retained identical file hashes.
PASS: The full guest device/driver inventory stayed unchanged. All 148 host OEM
INF names/hashes matched the baseline; no host cleanup was run for this release.
PASS: The native posted automatic administrator scan completed and exposed all
ten Windows estimates. The final standard-user run loaded that real report and
passed the fixed scan-to-results transition in the SWITCH native UI.
PASS: 26 final app/test Python source files were read independently from
the shut-down VM disk and matched this release. Fixture deletion/preservation was
also independently verified from that disk. Evidence JSON is in validation/.
PASS: All 37 EXE/DLL/PYD files match 0.7.1 and have valid publisher signatures.
The dedicated launcher remains byte-for-byte identical to PSF pythonw.exe.

WINDOWS-OWNED BEFORE / AFTER ESTIMATES
windows_updates: 199229440 bytes estimated before; 199229440 bytes remaining; remaining.
windows_delivery: 474314561 bytes estimated before; 0 bytes remaining; empty.
windows_setup_logs: 2424832 bytes estimated before; 0 bytes remaining; empty.
windows_chkdsk: 2424832 bytes estimated before; 0 bytes remaining; empty.

The remaining Windows Update estimate was disclosed, without an automatic repeat,
ResetBase or forced deletion. Successful cleanup does not imply every estimate
reaches zero. Windows estimates are not measured physical bytes deleted.

An initial VM run caught an oversized drive-enumeration buffer before cleanup.
It was corrected, covered by a real drive-inventory regression, and passed the
elevated engine rerun. The later UI check caught stale selections during the
results transition; a guard and regression test fixed it. The final standard-user
run verified that transition without repeating cleanup or requiring more UAC.
A canceled/failed elevation attempt was reopened using normal UAC; no protection
was disabled to continue.

No driver removal/restore/backup action was performed in this release's live test.
Those adapters and protections are unchanged; earlier validation is retained in
VM-VALIDATION-HISTORY.txt. One enforced VM validates this observed configuration,
not every future Microsoft reputation or policy decision.


SWITCH Clean 0.9.0 - recommendation validation
November 13, 2026

PASS: All 164 regression tests passed locally and in the disposable Windows VM.
The 29 new tests cover cleanup advice, verified remainder history, failure paths,
manual-only categories, engine integration and unchanged confirmation requirements.

PASS: The native portable host launched normally in the VM with Smart App Control
Enforced before and after. No Run anyway, SAC policy changes, exclusions, download
mark removal, own signing, installation, or UAC prompt was used for this test.
All 37 executable/runtime binaries retain valid publisher signatures and are
byte-identical to 0.8.0. The launcher is the unchanged PSF-signed pythonw.exe.

PASS: A controlled saved-scan fixture exercised the real native interface and
local history loader. The 380 MiB Windows Update remainder displayed Leave for
Windows, 64 KiB Delivery Optimization displayed Small amount, shaders displayed
Keep, and Brave remained optional and unchecked. The summary said nothing was
worth cleaning in routine categories. Clean selected was disabled with no choices.
Use suggested selection left all items unchecked. No cleanup operation was run.

PASS: This computer's saved November 12 administrator scan was independently
replayed against the new advice engine. Its verified Windows Update remainder was
imported into the private local history for this Windows build/current volumes.
The now-disconnected F: volume had a zero prior estimate and was omitted from
the imported scope. This was historical replay, not a new host cleanup or scan.

PASS: Final app/test source hashes were independently read from the shut-down
VM disk and matched the release. Evidence JSON is stored in validation/.
Guest OEM driver INF records remained unchanged. All 148 host OEM driver INF
names/hashes matched the preservation baseline. Brave, Razer and THX were untouched.

The cleanup/driver adapters, exclusion rules and native theme are unchanged from
0.8.0. Live destructive cleanup tests were not repeated for this advice-only
change; earlier validation is preserved in VM-VALIDATION-HISTORY.txt and history/.
One observed SAC configuration cannot guarantee future reputation/policy decisions.
The 100 MiB combined / 1 MiB per-item suggestion thresholds and 30-day history
lifetime are application policy, not Microsoft requirements or promised savings.


SWITCH Clean 0.9.1 - missing scanner lock validation
November 13, 2026

PASS: All 183 regression tests passed locally and in the disposable Windows VM.
The 19 new tests plus the updated missing-lock regression cover read-only scans,
missing-lock recovery, concurrent creation, busy locks, access/write failures,
unsafe ancestors, linked files, current-folder pointer changes, modified cleanup
candidates, exception release and the native report. Real Win32 file/locking APIs
operate on disposable fixtures; selected faults and race conditions are simulated.

PASS: The portable native app launched normally with Smart App Control Enforced
before and after the VM run. No Run anyway, policy changes, exclusions, download
mark removal, installation, own signing or UAC prompt was used in this test.
All 37 executable/runtime binaries retain valid publisher signatures and are
byte-identical to 0.9.0. The launcher is the unchanged PSF-signed pythonw.exe.

PASS: With only the inventory roots redirected to disposable VM fixtures, a real
Engine scan left the missing refresh.lock absent. Confirmed Engine cleanup then
created and acquired the missing lock, revalidated the scanner pointers and
deleted the two selected aged test files. The one-byte coordination file was
retained, and the native activity report showed the recovery explanation.
Servicing, process checks, file snapshots and deletion safeguards were not bypassed.

PASS: A second real cleanup attempt, while a competing process-compatible byte
lock was held, preserved the scanner test file and reported coordination blocked.
Seven protected fixture files remained byte-identical, including current scanner
files/pointers and Brave, Razer, THX and NVIDIA test data. No driver operation ran.

PASS: After normal VM shutdown, an independent read-only inspection of its disk
verified the deleted and preserved files, retained lock, result records and all
29 app/test source hashes against this release. Guest OEM driver INF records were
unchanged. Evidence JSON is included in validation/.

PASS: All 83 tracked host scanner files/pointers/coordination metadata and all
148 host OEM driver INF names/hashes matched the pre-work preservation baseline.
No cleanup was run on this computer. Its current scanner folders, Brave, drivers,
Razer and THX were left alone. Only isolated local and VM test fixtures were used.

Driver cleanup/restore/review, reinstall, recommendation, safe-file-deletion and
theme modules are byte-identical to 0.9.0. The broom icon and private local focused
selection/history are preserved. Private selections/history are excluded from the
portable ZIP. Earlier validation is retained in VM-VALIDATION-HISTORY.txt and
validation/history/. Other Windows cleanup categories were not rerun in this test.

This documents one observed SAC-enforced VM configuration, not a guarantee of
future Windows reputation or policy decisions. No Windows security setting changed.


SWITCH Clean 0.10.0 - saved scans and list ordering validation
November 13, 2026

PASS: All 201 tests passed locally and in the disposable Windows VM. Eighteen new
tests cover roundtrips, legacy administrator/CLI scans, validation and size limits,
privacy filtering, plan invalidation, execution rejection, save failures, cancelled
picker, saved selections, fresh-scan action routing and stable availability ordering.

PASS: The portable native host launched normally with Smart App Control Enforced
before and after the VM run. No Run anyway, security-policy changes, exclusions,
download mark removal, installation, own signing or UAC prompt was used for this
test. All 37 executable/runtime binaries retain valid publisher signatures and
are byte-identical to 0.9.1. The launcher remains the unchanged PSF pythonw.exe.

PASS: In the VM, an older scan was opened through the built-in native App.open_scan
flow. The saved date/version and read-only mode appeared, all selections were empty,
selection helpers could not check items, and the only primary action was Scan again.
The engine had no executable plan. Selectable rows appeared first in stable order.

PASS: A real standard-user Engine scan, with only inventory roots redirected to
isolated VM fixtures, completed through the native background/event flow and saved
its results automatically. Reopening that saved file preserved its timestamp,
cleared selections and discarded the executable plan. No cleanup was invoked;
the aged fixture file and guest driver records stayed unchanged. Administrator
Windows estimates were not rerun for this display/persistence change.

PASS: After normal VM shutdown, independent read-only disk inspection verified
the saved scan, preserved fixture, result records and all 31 app/test Python source
hashes against this release. Evidence JSON is included in validation/.

PASS: On this computer, the release executable opened normally with SAC Enforced.
More > Open saved scan... opened the Windows file picker and loaded the previous
0.9.1 administrator results. Visual and accessibility inspection confirmed all
20 rows, original date/version, 0 selected / 0 B estimated, read-only mode, Scan
again action and available categories before empty rows. Windows Update retained
Leave for Windows, while Brave remained Optional and unselected.

PASS: All 83 tracked host scanner files/pointers/coordination metadata and all
148 OEM INF names/hashes remained unchanged. No host scan, cleanup, driver action,
drive optimization or reboot ran while building/testing this release. Current
scanner folders, Brave, Razer and THX were left alone. Temporary unit-test fixtures
are isolated from user data. The prior administrator scan remains available locally.

The cleanup adapters, scanner exclusion/locking, safe file deletion, driver checks,
backup/restore, recommendation engine, theme and broom are unchanged from 0.9.1.
Private local selection/history are preserved. scans/, reports/ and local preference
files are excluded from the distributable ZIP. Previous evidence is retained in
VM-VALIDATION-HISTORY.txt and validation/history/.

This documents the observed SAC configurations; future reputation/policy decisions
can differ. No Windows security setting was changed.
