test: prove clean-machine onboarding and stable-staging isolation

[bradtaylorsf]

Summary

Run final acceptance from clean supported macOS and Windows environments and prove the complete install, onboarding, integration, recall, desktop, and isolation experience.

Why This Matters

Unit and component tests cannot prove that a non-technical user can go from a pasted prompt to durable cross-chat memory without risking their stable data.

Proposed Approach

• Use clean Apple Silicon 16 GB and Windows 11/WSL2 8 GB+ GPU environments.
• Execute the canonical prompt without repository-local assumptions.
• Verify preflight, installation, desktop launch, first import, folder preview, organization proposal, client wiring, and cross-chat recall.
• Run a destructive staging test fixture and prove stable database/archive hashes and service state do not change.
• Exercise upgrade, repair, rollback, uninstall/reinstall, cancellation, and failure recovery.
• Capture desktop and 375px visual evidence and verify package checksums.

Acceptance Criteria

• Both supported platform baselines complete zero-to-first-memory onboarding.
• Every selected supported client passes a real MCP or CLI recall check.
• Stable state remains unchanged throughout staging development and mutation tests.
• No login or manual Answer Engine API-key creation is required for the local UI.
• Background services recover after reboot and Electron remains only the control plane.
• All child acceptance criteria, learning files, merged PRs, and issue checklists are complete.
• pnpm verify and final installer/package verification pass from clean checkouts.
• No paid enterprise code is present in OSS and no npm package was published.

Out of Scope

• Adding features discovered during acceptance; file linked regression issues instead.
• Weakening checks to accommodate platform failures.

Related

• Parent epic: epic: deliver zero-to-first-memory desktop onboarding and isolated local channels #40
• Depends on: feat: isolate stable and staging local Answer Engine channels #41, feat: add a cross-platform one-prompt installer and capability preflight #42, feat: install and verify Answer Engine skills, plugins, MCP, and CLI wiring #43, feat: guide and reconcile the first agent-history import #44, feat: add permissioned preview-first local folder ingestion #45, feat: propose and apply reversible AI-assisted memory organization #46, feat: prove first memory with a grounded cross-chat recall tutorial #47, feat: ship a lightweight desktop launcher and local release controls #48, fix: adopt legacy stable installs and open the real desktop web runtime #68, feat: ship immutable installer downloads and dependency-aware bootstrap #71, fix: keep local client credentials out of process arguments #74, docs: make OSS licensing and community policies unambiguous #76, security: decide public Git history and contributor identity policy #77, fix: remove unsupported GHCR visibility mutation from release workflow #78, security: harden GitHub settings for public contributions #79, fix: patch dependency advisories found by public-release audit #86, fix: avoid full-history archive churn on incremental transcript updates #88, security: rebuild the public runtime from a patched base and dependency graph #89, release: make the OSS repository and runtime image public #80, release: publish and anonymously verify v1.1.2 installer assets #81, test: validate v1.1.2 on clean Apple Silicon macOS #82, test: validate v1.1.2 on clean Windows 11 with WSL2 #83
• Keep Graphify excluded.
• Do not publish npm.

[bradtaylorsf]

Acceptance plan:

1. Inventory already-provisioned clean Apple Silicon and Windows 11/WSL2 GPU environments without downloading VM images or touching stable memory.
2. Reconcile every feat: isolate stable and staging local Answer Engine channels #41–feat: ship a lightweight desktop launcher and local release controls #48 acceptance record, merged PR, learning file, and epic checklist entry.
3. Run the clean checkout, installer/package, onboarding, client recall, reboot/background-service, and destructive staging isolation matrix on each genuinely supported available baseline.
4. Re-run final repository/boundary verification and the epic verify-only audit.
5. Record exact evidence. If a real Windows 11/WSL2 8 GB+ GPU baseline is unavailable, do not infer success: document the environment blocker and leave test: prove clean-machine onboarding and stable-staging isolation #49 and epic: deliver zero-to-first-memory desktop onboarding and isolated local channels #40 open as required.

Disk guard: no VM or OS image downloads; generated package output remains bounded under the issue worktree and is removed after inspection.

[bradtaylorsf]

Verification status and Lesson Learned

The exact required audit node ../alpha-loop/dist/cli.js run --verify-only 40 completed with verdict PARTIAL. Its operational findings match the clean-environment gate: no real concurrent dual-Compose run, no published immutable v1.1.0 installer/tag, no clean supported Apple Silicon or Windows installation, and no recorded live Codex/Claude cross-chat integration run. These cannot be inferred from fixtures and remain open.

The verifier also reported #46 as unmerged even though PR #64 merged at a37ceed941973f998179ad2b66d9a9f89f5a668d and #46 is closed. Root cause: the merged PR body used Related to #46 instead of a closing reference, so verifier association missed it. PR #64 metadata now says Closes #46; the issue-to-PR GraphQL relation is the regression check. Future child PRs must audit both baseRefName and issue closing-reference association before merge.

[bradtaylorsf]

Follow-up on the #46 verifier association: closingIssuesReferences stays empty because child PRs merge into a non-default epic session branch, but the issue timeline now contains a merged PR #64 CrossReferencedEvent. That timeline association—not default-branch auto-close—is the correct regression check for session children.

Exact external blockers (2026-08-15 America/Los_Angeles)

• Local host: Apple M5 Max / 128 GB, but it is the active personal machine and therefore not a clean baseline. No Parallels, UTM, VMware, VirtualBox, Multipass, Lima, or Colima environment is installed. Only Docker Desktop exists.
• Repository Actions runners API: total_count: 0; there is no clean self-hosted macOS or Windows runner.
• GitHub-hosted ARM macOS: standard runners provide 7 GB and larger M2 runners 14 GB—both below the agreed 16 GB baseline—and macOS runners do not support nested virtualization, so the Docker runtime cannot be exercised there.
• Windows: no provisioned Windows 11/WSL2 host with a supported 8 GB+ GPU exists. Standard hosted Windows is not that baseline; provisioning a paid GPU runner or multi-gigabyte VM is outside this task and was not inferred.
• Canonical release inputs: npm view @answer-engine/create@1.1.0 returns E404 and refs/tags/v1.1.0 is absent. The goal forbids npm publication, so I did not create either.

Environment-independent evidence remains green: fresh session checkout install reused cache with zero downloads; pnpm verify passed all 534 tests/builds with one environment-gated DB skip; the public-boundary check passed; raw archive remained 0 B; 1.1 TiB remains free. #49 and #40 stay open.

[bradtaylorsf]

Draft blocker handoff PR #67 now has both independent CI jobs passing (verify and container). It targets the active epic session branch, remains draft, and intentionally does not close #49. The local issue worktree and its 436 MB dependency tree have been removed.