[mikecerisano]

Check could report preserved Finder metadata as missing, and two selected subfolders on one Mac disk could miss physical-drive resolution. This fixes both, preserves destination progress through MHL/report work, bounds diagnostic export reads, and corrects PDF pagination and preview counts. It also removes quadratic result lookup and a repeated queue-view update warning during many-file transfers.

Two small additions stay tucked away: reviewed transfer-preferences import/export in Settings, and a Mac Advanced hint for backups reporting a USB 2 connection. Verification defaults, source protection and overwrite rules remain unchanged.

Most of the diff is regression coverage and reproducible local harnesses. Completed checks include:

• Baseline engine Debug/Release, each: 188 XCTest passes / 41 skips and 217 Swift Testing passes / 6 skips; zero failures. Full Mac: 1,282 passed / 23 skipped. iPad/iPhone: each 75 passed / 10 skipped. Universal Mac and iOS-device Release builds passed.
• Final combined source with Handle ambiguous filenames without crashing transfers #63 and Continue queues after reviewed metadata exclusions #64: each engine configuration passed 205 XCTest / 43 skipped and 217 Swift Testing / 6 skipped; full Mac 1,301 passed / 23 skipped; iPad/iPhone each 75 passed / 10 skipped. Runtime-warning checks were clear for the accepted native runs. Earlier mobile controller timeouts remain documented separately.
• 636 mounted transfer/policy scenarios in qualified split runs, 60 recovery scenarios, and 68 filesystem-fault passes / 4 expected skips. FAT32 maximum-file and oversized-refusal cases passed without replacing prior media/history.
• Physical SD to two SSDs: nine mode/pipeline/reuse scenarios through MHL and Saved Check, with source hashes preserved.
• Two 119.6 GB large-file runs passed: physical exFAT SSD to SSD, and exFAT SSD to a journaled HFS+ image backed by the other SSD. Both observed two large verification readers and a 32 MiB sampled engine peak. HFS+ reused permitted readback evidence for MHL; exFAT still rereads. These are earlier recorded engine revisions, not GUI memory or a physical HDD/dock reproduction.
• Actual coordinator: 20,000 files to two backups; separate 2,000-file PDF runs checked 4,000 entries across 144 pages. Generated 4K H.264, HEVC and ProRes previews, repeated reports, 256 seeded soak iterations, a 20,009-file engine case and 20 process-death recovery scenarios passed at their recorded revisions.
• Selected Thread/Address Sanitizer and translated x86_64 checks passed. A 100,000-row host History probe retained its evidence; it is not an app-search or leak proof.
• Seventeen Python cleanup regressions passed, including two cases observed failing against the older helper. Physical test folders were removed after compact evidence preservation and normal image detach. Six shutdown test simulator clones remain while unrelated project builds are active.

Full revisions, exact counts, corrected harness attempts and limits are in the overnight validation record. A local signed universal diagnostic archive is retained; it was not published. Native final import actions, progress/status-menu walkthroughs and physical mobile-provider/background behavior remain unvalidated. The paired application benchmark stopped at fixture setup before any transfer, so it supplies no comparative performance result. GitHub Actions credits are exhausted; these gates ran locally.

Issues #10 and #23 stay open. The tests do not establish either reporter’s exact failure mechanism. No merge or release is included. Filename safety is separately reviewed in #63, followed by the reviewed-exclusion queue policy in #64; updater metadata is independent in #62.