Observed development checks:
cargo build -j 2: passed on x86_64-pc-windows-msvc.cargo test -j 2: 45 tests passed (28 core integration, 11 advanced integration, 6 native workflow tests), plus library/binary/doc test targets. No tests were ignored.cargo fmt --check: passed.cargo clippy -j 2 --all-targets -- -D warnings: passed.- Native executable launched and rendered real overview, firmware, flashing, root, utilities and settings pages to PNGs in ../artifacts/. Screenshots were inspected; the initial theme contrast defect was corrected.
cargo build --release -j 2: passed after final safeguard repairs.- Packaged
dist/PixelFlasher.exepassed real executable inspection, nested ZIP extraction, persistent-library and read-only legacy-import smoke checks. Results and binary SHA-256: ../artifacts/executable-smoke.json. - Strict Akinator knowledge coverage: exit 0, no findings; report in ../artifacts/knowledge-coverage.json. This checks documentation structure, not parity.
- The final packaged release also launched and rendered its real overview page.
- Automatic approval review blocked cleanup of one 8,192-byte development fixture:
%TEMP%/pixelflasher-smoke-x_grqdcy/PixelFlasher4.db. It contains synthetic records only and was left in place. The smoke script now explicitly closes its Python fixture connection, and subsequent smoke runs cleaned up normally.
Core tests exercise multi-device/mode/property parsing, serial-bound plans, locked-target rejection, input changes after review, stopped sequences on failure, slot selection, known-bad SDK releases, boot headers, archive traversal, nested firmware extraction, OTA hashes and delta rejection, persistent image provenance, read-only legacy imports, shell identifiers and redaction. The executor also rejects mismatched/missing/overriding command target selectors before any transport call. Personal tool command arguments and known echoed argument values are omitted/redacted at log creation.
Advanced tests cover certificate metadata/issuer verification/revocation lookup, absence of private-key material in reports, root archive traversal boundaries, SU policy identifiers, root configuration paths, support ZIP redaction and native root-patch plans that do not flash.
Not validated: physical device detection, flashing, root patch execution, bootloader locking/unlocking, OTA recovery completion, module changes, wireless pairing, scrcpy, downloaded firmware catalogs, online certificate status access, or non-Windows packaging. No attached phone was flashed or otherwise changed. The native screenshots use an isolated data directory and contain no fake device.
Fixture tests are evidence about code paths, not an assertion of full feature parity or hardware safety. See PARITY.md for remaining conversion work.
Independent read-only librarian review initially blocked target-selector enforcement and personal-tool argv redaction. Both were repaired with regression tests. The final reinspection returned CLEAR for those repairs and documentation truthfulness; it did not run hardware operations or certify full application parity.
Regression coverage verifies diagnostic output isolation from editable root configuration, persistent root preference, selected-serial retention after scans, disabled-device filtering and stale package-selection clearing. A worker test proves plan preparation opens review without executing its device command. JSON5 tests preserve nested/hex values and reject invalid, duplicate or non-finite values. AVB tests extract a property descriptor and reject overflowing descriptor lengths. Platform readiness requires both executables, not merely a directory.
UI captures for this batch are in ../artifacts/v02-overview.png and the other v02 page captures. Final release smoke results remain in executable-smoke.json. No device commands were executed by the development tests or capture workflow.
Live debug captures exposed duplicate scroll-area IDs in the firmware view. All scroll areas now have distinct stable IDs; a native rendering regression checks all nine pages for painted collision warnings. The initial screenshot showing the defect was replaced after repairing and re-rendering that page. Version text in support ZIPs and network user agents uses Cargo package metadata.
Workspace suite: 72 tests (9 native UI, 11 advanced, 28 original core, 24 Studio). Original 45 tests are retained. Added evidence covers immutable digest approval, secret serialization/redaction, six pinned version pairs, classic/V2 restrictions, UID/slot mismatches, critical checkpoint stop behavior and actual module-install plans for all providers. SQLite migration fixtures preserve legacy rows and make backups; interrupted writes require review, never replay. Backup fixtures reject changed content/device/build/slot and checksum mismatches. Resume range and catalog row association parsers reject malformed/mismatched inputs. This is not a real network download/resume test or hardware boot-image compatibility validation.
The renderer exercises all nine workspaces in empty/populated/busy/failed/expert states, both themes, at 1024x720 (90 renders). Existing all-page widget-ID warning regression remains. Render tests are not complete keyboard/accessibility or visual approval for every populated state. Native empty-page release captures, packaged CLI and stdio smoke evidence are recorded separately in studio-validation.json.
Independent librarian review found two defects before release: free-form MCP metadata leaked sensitive labels/descriptions, and Magisk --install-module was missed by a substring classifier. Metadata exports now use generic/allowlisted fields; Plan::add explicitly marks destructive privileged stages critical. Both have regression coverage; final read-only reinspection returned CLEAR. Matching restores freshly revalidate Android build before entering fastboot; bootloop-only restoration is still unsupported.
No real phone scan, flashing, rooting, module change, backup/restore, relock or wireless pairing was performed. No hardware-verified combination is claimed. All remaining 1.0 gates remain in IMPLEMENTATION.md and PARITY.md.
Final UI coverage uses the production draw method, including activity/error UI. Provider manager-install fixtures cover synthetic manifest metadata for all twelve pinned pairs and reject a mismatched selected provider. They do not validate APK signatures, official binary assets, kernel interfaces or installation on a phone. Configuration read results reject changed serial/source paths, and target changes discard pending reviews and sensitive drafts. Frozen review summaries include lock/unlock wipe semantics and explicit set_active slot transitions.
Adds three regression tests: default-false/legacy-settings consent and correct factory/OTA cookie selection; missing acknowledgement rejected before network; current Google factory/OTA row fixtures retain their checksums. Fixture metadata comes from https://developers.google.com/android/images and /android/ota, fetched with their respective acknowledgement cookies. Full pages are development artifacts only; the source package includes two small table rows, no firmware payloads. Live Rust executable factory/OTA retrieval, release hashes and native firmware-page captures are recorded in catalog-validation.json. No phone operation occurs.
Final packaged v0.3.1 verification passed: 75 workspace tests, formatting, strict Clippy, release build, original CLI smoke and MCP smoke. The Rust executable retrieved 2,359 factory and 2,121 OTA records, all with SHA-256 values. Native firmware-page captures in both themes show the acknowledgement control and links. Independent source/document review returned CLEAR. Counts are a live snapshot, not a permanent catalog size guarantee; no firmware was downloaded or flashed.
The device bar is drawn once above the page, with Scan and Inspect. Overview no longer paints the marketing panel or the four-step essay; its three status rows open Update Studio, Firmware & Images, and Settings. Google catalog rows are grouped by the filename model and listed only for the chosen model. Boot images, applications, modules, and partitions use the same table layout. Applications and verified backup sets stay multi-select because those actions apply to a set. Reboot, flash, lock, wipe, and patch stay on their pages and still open a reviewed plan.
cargo fmt --all was applied. cargo test --workspace -j 2 passed (77 tests), including catalog_file_names_split_into_model_and_build, catalog_table_lists_only_the_chosen_model, the nine-page widget-id check, and both themes at 1024x720. cargo clippy --workspace -j 2 --all-targets -- -D warnings passed. Native captures of the empty overview, firmware, applications, root, and backups pages were taken from the debug binary with an isolated PIXELFLASHER_DATA_DIR and inspected. A nested grid inside a scroll area painted a black frame; those lists use fixed columns instead. No phone operation was performed.
The shared core relocation and branding changes pass all 75 workspace tests, formatting, strict Clippy and Windows release builds. Both packaged binaries pass the existing CLI/MCP smoke checks. Live factory/OTA metadata retrieval still works; counts and executable hashes are in release-validation.json. The native Settings page was captured from the branded release binary and inspected. No phone operation was performed. Public-file scanning found no private PEM keys or GitHub/AWS token patterns; public certificate fixtures remain included.
Historical binaries/source packages are preserved locally under dist/archive. Windows CI is configured separately and does not certify hardware behavior. Versioned portable/source archives exclude data, cached pages, build directories and historical binaries. Release publication/remote CI evidence is separate from these local checks.