Selected service and integration qualification
Prepared November 12, 2026, for the VCC Plus technical paper. This is an authored, selected summary of retained development qualification, not a new test run, independent replication, complete execution log or claim that the evolving public branch exactly equals every measured build. The original records remain outside the publication payload. Public source and fixture links below permit static inspection and future reproduction with separately supplied inputs. Counts refer to their stated selections; overlapping checkpoints must not be added together.
Transactional disk-image editing
The November 10, 2026 editor record reports 55 selected backend, safety, insertion, persistence and hard-disk tests passing on arm64 and separately on x86_64 under Rosetta on macOS 27.2 / Xcode 27.2. Safety/insertion checks exercise both guest processors. Native interaction checks covered DECB and OS-9 import/export, private edits, canceled and confirmed restart/apply, reopen, Undo/Redo and byte-for-byte discard. Supplementary DECB bytes survived apply/reopen unchanged. Descriptor-preserving OS-9 moves, hard links, directory growth, full-disk rollback, stale requests and whole-image Undo/Redo have authored fixtures. A native drag moved a file and preserved its exported bytes; synthetic edge-drag resizing was not reliable enough to claim verified interaction.
The November 11 hard-drive extension is a separate checkpoint: 67 selected checks passed per host slice with no skips, comprising 22 filesystem/backend checks, three working-store checks and 42 bridge/model/media safeguards. They cover allocation clusters, numbered HDB-DOS regions, bytes outside the edited region, Undo/Redo, protected/temporary media, stale leases, aliases/external changes and canceled/confirmed application. Applicable execution and locking checks cover both guest processors. Three isolated native cases exercised hard-drive volume selection/apply, DECB round-trip/cancel and supplementary-track preservation. A separately supplied combined OS-9/HDB-DOS image also round-tripped in a working copy; its source was unchanged. That observation is filesystem-editor evidence, not an OS-9 boot qualification, and the private image is not distributed here.
Backend, backend fixtures, safety fixtures and ToolShed provenance identify implementation and upstream filesystem-code attribution.
Guest SDK and compatibility subset
The November 7 library qualification includes 28 bank-transfer guest sessions. It compared the entire physical reservation and adjacent guard bytes with an independent expected image, exercised 513-byte transfers across all three internal bank boundaries, rejected invalid ranges, and checked restored mapping and IRQ/FIRQ masks. Standard 512 KiB, 2 MiB and 8 MiB configurations accepted transfers; 128 KiB, disabled MMU, alternate task and an unsuitable data layout refused the reservation. These configurations use the documented BASIC mapping; they do not establish support for arbitrary user MMU layouts.
Other selected library records report four heap-example runs across two optimization levels and both guest processors; GIME visible-page/palette cleanup checks; and writable-stream checks on Disk BASIC, RGB-DOS and SDC-DOS paths. Refused cross-linked/open-BASIC-file writes left full images unchanged. The positioning selection covered fragmented 65,537-byte files, sector/granule boundaries, text bookmarks and protected dirty-seek retry; its 18 sessions were recorded separately from earlier read/write selections. No aggregate number of unique library tests is asserted. None of these records establishes a complete ISO C/POSIX runtime or physical-device qualification.
Bank fixture, GIME fixture, background audio fixture, heap fixture and writable stream fixtures retain the bounded procedures. Guest SDK integration must remain distinct from CMOC, BGraph/BControl, WydeAlloc and other upstream algorithms and notices.
Content-independent source builds
The November 12 shared build-profile qualification reports 57 focused Python checks, including profile/schema/safety, public dependency preparation, reader/runtime resources, template paths and existing catalog packaging. No new dependency download occurred; verification inspected existing inputs. Normal-profile checks compared retained resources with prepared inputs. Source-profile checks omitted full manuals and used invalid maintainer-path sentinels to establish that those paths were not consumed.
Normal arm64 qualification initially failed an assertion demanding a repository ignore file in a downloadable project; that assertion was narrowed to the intended executable/resource/build membership, and the focused rerun passed. The other 51 selected cases passed unchanged; one additional native collection case passed. The normal x86_64/Rosetta selection passed all 55 cases together. The source arm64 52-case selection passed with 24 expected skips: eight full-manual and 16 optional-compiler cases. Additional Assembly/collection checks passed separately. These are differing selections, not interchangeable all-platform totals.
A universal source Release build passed, and 67 ROM-free Release core checks passed on each compiled host slice. Source resources had zero software titles and Archive documents, lookup help/runtime prompts, no StarterKit/ROMs and no optional compiler payload. Source UI tests compiled for testing; no desktop interaction was run for this change. This shared-profile record is separate from the OSS export prototype qualification and does not imply another export or runtime test merely from selecting the profile.
Profile implementation, lookup generation and explicit dependency preparation provide the public implementation. Preparing dependencies and compiling offline are distinct actions. ROM-dependent reproduction requires the user's own inputs.
Mission Recorder integration
The retained Mission Recorder entry is part of the November 2026 C-workspace qualification; the entry does not independently timestamp every flight. Its FD-502 matrix reports seven flight variants, two optimization levels and two guest processors (28 flights), plus four alternate-storage boots (32 flights total). Variants include direct/disk launch, write-protected retry, silent and cartridge-sound combinations, missing requested devices, and discard.
In the marked-sector probe, the pilot selected sector 15 and then steered to sector 06. Physical address $50000 retained 15, while independent disk extraction and serial printer capture matched the same 27 report bytes. A protected write left disk and printer unchanged; retry in the same flight after protection was removed saved LOG03.TXT and preserved LOG01.TXT and unfinished LOG02.TMP. Release disk boots under RGB-DOS and SDC-DOS on each processor produced the expected sector-07 report and returned to a usable BASIC prompt. Separate 128 KiB launches refused the memory requirement and preserved the disk.
With both audio devices installed, speech-chip advancement and continued stereo DAC sample delivery were observed while steering. Fixtures compared the IRQ vector, MMU mapping and BASIC palette before/after the flight, checked playback stopped, and evaluated a fresh BASIC calculation. Printer evidence is guest serial output captured by the emulator, not physical printing or host-printer acknowledgment. The report is an integrated functional case, not a hardware benchmark or complete gameplay/analog-audio qualification.
Example source, qualification orchestration, FD-502 fixture and alternate-profile fixture retain the authored program and procedures. These fixtures may require separately supplied system ROMs and a prepared toolchain; their presence is not an execution receipt for the current public branch.
Common limitations
All host-slice execution above was on Apple Silicon; x86_64 used Rosetta. Physical Intel and macOS 14 runtime were not qualified by these records. Authored filesystem tests do not establish a full OS-9 system boot. Digital sample/capture checks are not a human listening study. Public source inspection, recorded execution and future independent reproduction remain separate evidence.