Swift versus stock Qwen3.8 Apple feasibility campaign

Decision: unresolved, no promotion. No candidate was downloaded, loaded or queried. The campaign reached an evidenced safe-trial barrier; the model has not failed a quality test. The exact usefulness corpus was not frozen and all performance/quality/capacity/vision/stability gates remain missing.

• Identity and exact artifact bytes
• Dated source registry
• Run plan and thresholds
• Requested coverage and gaps
• Planned workload, not an executed corpus
• Feasibility input and deterministic result
• Optimistic disk-policy screen
• Managed runtime validator result
• Host baseline and hardware
• Friction and durable follow-up
• Restoration boundary
• Decision summary
• Publication summary
• Hashed role ledger

The 16 GiB reserve leaves a nominal 32 GiB candidate envelope, not a measured Metal working set. Memory and media overhead remain unbounded. Artifact file sizes are not resident allocation measurements. The disk screen retains 10 GiB plus a 1 GiB runtime/evidence allowance and already fails without temporary duplication.

There are zero matched timing cells, so no chart or efficiency metric can be derived. The public result describes inventory, source metadata and pre-load readiness only. Native snapshot inspection verifies local links, not upstream artifact completeness or inactive-service ownership.

Redactions: host and account identifiers, private addresses/ports, paths, service labels/PIDs and deployment assignments are omitted. Public hardware, storage, policy and model artifact identities remain.