# Release-candidate audit

DiskGolf 0.1.0-rc.2, September 8, 2026. This audit distinguishes demonstrated behavior from remaining release work. Public publication has not occurred.

| Requirement | Evidence | Status |
|---|---|---|
| Original visual identity and three concepts | Midnight, Daylight, Observatory renders; visually inspected dark/light/compact layouts | Implemented and reviewed locally; user appearance feedback remains welcome |
| Native folder/drive scanning without administrator privileges | Installed application scans known fixture through its real interface; process token is not elevated | Passed on reference host |
| Logical and allocated totals | Native tests for known trees, hard links, sparse files, deep paths and Unicode; installed fixture matches independent byte count | Passed |
| Cancellation and partial results | Million-entry benchmark, native cancellation and partial-state checks | Passed on reference host |
| Links, offline entries and access failures | Junction loop avoidance, offline-attribute fixture, explicit denied-folder ACL, exclusively locked file | Passed; live cloud-provider compatibility is not claimed |
| Navigation, sorting, filters, empty states | Browser integration and packaged UI tests; native multi-page ordering and stale scan checks; RC2 duplicate-name chart selection regression | Passed |
| Full paths, percentages and allocation context | Selected-item details, logical/allocated mode, whole-drive accounting-gap display | Implemented |
| Explorer reveal and clipboard | Packaged path copy and native Explorer command pass; Shell selection independently matches the exact fixture file | Passed |
| Reviewed recycling and restore | Real UI review, native Windows recycle, Shell restoration; HTML companion directory preserved | Passed on disposable fixtures |
| No permanent fallback; stale changes rejected | Locked-file recycle fails with original intact; changed-file rejection; root/folder rejection; duplicate IDs and hard-link allocation checks | Passed on reference host |
| Diagnostics exclude paths | Packaged download; exact seven-field schema asserted | Passed |
| Keyboard, contrast, reduced motion, scaling | Browser keyboard integration, packaged WebView2 axe WCAG A/AA checks including dialogs, reduced motion, compact and 2x pixel-density renders | Passed locally; independent Windows scaling check remains |
| Large datasets and rendering | 10k/100k/1m native benchmark; packaged 1m test; 500-path SVG/canvas comparison | Recorded in performance.md |
| Installation and uninstallation | Final-candidate current-user NSIS install/uninstall in isolated directory and WebView2 profile; portable ZIP launched and scanned; executable and registry entry removed; scanned file preserved | Passed on existing host; fresh Windows/second-machine check outstanding |
| Dependency licenses and name check | 278 dependency/installer/SDK records, no unknown declared license or missing notice text; Microsoft WebView2 loader matched to official SDK archive; preliminary exact-name search | Notices assembled and installed copy verified; working-name collision documented |
| Distribution artifacts | Installer, portable/source archives, SHA-256 hashes, notices, privacy statement, release notes and local landing page | Local artifacts assembled; see SHA256SUMS.txt |
| Signing and public feedback route | Candidate is unsigned; owner-controlled publication/account/signing steps documented | Public launch pending; no accounts created or money spent |

## Outstanding public-release gates

- Independent Windows 11 installation, display-scaling and basic file-action verification using beta-checklist.md. No pristine Windows machine or VM is configured in this workspace.
- User review of the actual visual result and a small real-user beta. Automated tests cannot establish whether new users understand the interface without instruction.
- Publisher/distribution decision and a real public feedback destination. No signing identity, domain, public repository or download host has been configured by this build.

The candidate can be evaluated locally. These remaining items must not be represented as completed or as a guaranteed public-release outcome. See the accompanying package checksum for the exact binary being tested.

## Candidate identity

RC1 is preserved with its own checksum under release/history/0.1.0-rc.1. The tested RC2 installer is DiskGolf_0.1.0-rc.2_x64-setup.exe, SHA-256:

`a8641346eaac1a0e98e01ea0a5c94cbd5de2f4b263faa345f98477b6150befb6`

## RC2 change verification

The chart/model code now carries a containing-branch identifier for linked hover. A regression uses duplicate filenames from different sample folders and an actual pointer click on the outer chart ring, then checks the precise selected path and review-list path. The UI requests exact metadata by scan/item identity, resets the containing folder's list to size order, and rejects late responses after navigation or a replacement scan. The previous filename-search fallback was insufficient evidence for exact chart selection.

Review entries retain their originating scan identity. A delayed-response browser regression verifies that an old view cannot replace new results or queue a stale file during a scan switch. Installed-native checks reject recycling against the old scan even when the current scan reuses the same numeric item ID; both fixture files remain intact.

The current source passed 13 native tests (12 library tests and one integration test), three model tests, and six browser integration tests. The new integration test renames a disposable scan location while traversal is active, verifies partial/skipped results, and preserves its files. This is not a physical removable-drive disconnection test.

The installer identified above passed fresh-process scan/navigation/clipboard checks, exact duplicate-name chart selection, pagination and stale-file/scan rejection, reviewed recycling and restoration, dark/light accessibility checks including dialogs, diagnostic export, and cancellation/invalid-path/empty states on the reference host. Its installed notices match the source notices byte for byte. The installed and standalone executables match after accounting for Tauri's three-byte installer-type marker.

RC1 artifacts and their audit remain under release/history/0.1.0-rc.1. Baseline million-entry performance measurements remain explicitly attributed to RC1 in performance.md; they were not rerun merely for the RC2 UI and packaging changes. The independent-machine and public-release gates above remain outstanding.