[wayri]

macOS target available for testing

Both 0.4.1 Mac candidates passed native packaging and installed-app checks from source 96963976f0131ab378ea6e34f289a3e254ce03d4.

Platform	Candidate download	Outcome
Apple Silicon	DMG + SHA-256	Native build, signed-member integrity, real circuit solve, installed/offline CLI, startup, reinstall and removal passed
Intel Mac	DMG + SHA-256	Same native and installed checks passed
Linux x86-64	Flatpak + SHA-256	Native build, hard-AS supervised circuit and installed lifecycle passed
Windows x64	No qualified matching-source artifact retained	Build/installed worker/CLI passed; frontend/native-panel proof timed out

Native workflow: https://github.com/wayri/SPIKE/actions/runs/40097197077 . Its overall conclusion is failure because Windows failed; aggregate package verification and publication were skipped. Successful Unix jobs are reported separately, not as an all-platform pass.

Explicit experimental Mac resource policy

The owner explicitly approved the experimental Mac RAM monitor. darwin_sampled_rss is selected explicitly for dedicated circuit-worker qualification; it is not a hard memory cap or app-wide limit. Strict API defaults remain unchanged. Bounded observation failures abort the job; terminal results, signed package manifests and DMG receipts state sampling, shared-page and descendant limits. macOS15 baseline, ad-hoc signed and unnotarized.

The first native attempt exposed Darwin EPERM for zombie-only groups. The correction accepts only bounded positive zombie evidence including the retained leader; live/missing/malformed observations retain the failure. Independent review found no consequential blocker. Actual native logs show33/33 tests on each Mac,29 passed/4 Darwin skips on Linux. Both Macs passed the offline-boundary test and actual installed CLI with networking denied.

Source and documentation

• Candidate code CI passed at9696397: https://github.com/wayri/SPIKE/actions/runs/40097076851
• README, platform guide, release notes and native evidence now include the candidate links and hashes, in documentation-only commit 88af7dca65a16823e457d219607c555876ec35c6.
• All759 Markdown files and the architecture guard pass. Code tests:50 locally,44 passed/6 native skips; the native cases are covered above.
• Checksum sidecars and ZIP directory sizes were retrieved through bounded ranges. Native CI verifies actual installed bytes; no full local Mac/Flatpak binary hash check is claimed.
• The shared dirty development checkout remains preserved. No concurrent numerical work was imported into the frozen candidate.

Remaining publication gates

Windows frontend/native-panel proof needs diagnosis without weakening exact PASS/normal-close acceptance. The whole app source is unchanged from its earlier passing Windows candidate; current evidence cannot distinguish pre-JavaScript startup, slow WebView or title forwarding. Next repair: retain bounded captured host output, title timeline and failed installer identities.

A complete matching-source package set, knowledgeable human numerical review and native desktop visual acceptance remain required. GUI process survival is not visual acceptance. Public release remains0.4.0-alpha.1; no0.4.1 tag or release was published.

Attribution

AI-assisted implementation and independent agent review follow the existing review requirements and do not replace human numerical review. All commits, DCO sign-offs and pushes use canonical wayri identity.