Retire adoption macOS CI and its obsolete selector - #957

Scope

Retire the three advisory adoption macOS jobs and their obsolete selector, retaining the Linux bootstrap and portable installer code. Enforce the exact Linux job/output contract, describe hosted receipts as historical, and provide a pinned restoration rehearsal verified in a disposable worktree.

• Base commit: eb31d233234cdc552fc81672087ac8769e3696e0.
• Lane: lane:foundation.
• Owned paths: adoption workflow, adoption manifest/guides, automation catalog/decisions/handbooks, affected contract tests, portability comments, and registered evidence.
• The landed resolver prerequisite fix(openhands): protect native shard discovery before macOS CI retirement #959 preserves ci_discovered for newly added root test modules. The Python 3.9 recording syntax guard remains executable.
• This publication contains the landing rebase plus one forward fix commit for the P2 and three P3 groups. The CC owns landing after the co-op range-diff micro and CC read.

SOTA sources

• Maintained pre-retirement main at a56f85e8397751d36fdbdc03910f15ac877ad00c: .github/workflows/adoption-bootstrap.yml, the removed CI test classes, and the landed Repair macOS validation portability and failure diagnostics #948 portable repairs. The restoration procedure reuses these source bodies and decorators.
• Landed resolver prerequisite #959 at 0a053e1e5861a5522dcbbeea1c5006c4d8937b95: derive_ci_protected recognizes native validate_shards.py run discovery; the own-repository test keeps its protected expectation.
• Git v2.53.0 git-restore and git-worktree: installed native restore --source and detached scratch worktrees bound the inverse.
• CPython v3.13.16 ast.py and AST documentation: AST locations select unchanged donor definitions; feature_version=(3, 9) provides best-effort syntax validation, with a rejected Python 3.10 match control.
• Current landed main eb31d233234cdc552fc81672087ac8769e3696e0: docs/lanes.md hot-file protocol, native handbook builder, host_receipts.register_file, generated-report tooling, and evidence normalizer. Main's manifest was taken wholesale and own files re-registered.
• GitHub Actions billing, free use and hosted runners, read 2026-10-10: public standard hosted macOS runners have no runner-minute charge; this decision reduces maintenance rather than claiming cost savings.
• CI-pinned kjanat/actionlint v1.17.0 and zizmor 1.30.1, matching the installed clients.

Evidence-class table

Local commands run

All substantive test/validation runs used nice -n 19 ionice -c 3. Full discovery belongs to hosted CI.

python3 -B -m unittest tests.test_workflow_hardening.AdoptionBootstrapLinuxPolicyTests tests.test_workflow_hardening.LinuxBootstrapChangesTests
# 13 cases, rc0; both new controls had failed before the policy fix.

python3 scripts/build_new_wsl_handbook.py --write
python3 scripts/build_new_wsl_handbook.py --check
# Both rc0; unchanged generated digests, both receipt histories preserved.

python3 -B -m unittest \
  tests.test_adoption_bootstrap tests.test_adoption_bootstrap_macos \
  tests.test_claude_harness_audit_bounds tests.test_claude_pr_review_workflow \
  tests.test_claude_pr_toolkit_review_workflow tests.test_codex_broker_reaper \
  tests.test_github_automation_practice tests.test_landscape_sweep_skills \
  tests.test_runtime_worker_openhands_push_gate tests.test_runtime_worker_openhands_resolver \
  tests.test_shell_parser_ci tests.test_skill_usage \
  tests.test_workflow_hardening tests.test_workflow_policy
# These 14 real modules contributed 1,233 cases, 34 skips, zero failures/errors.
# The original aggregate invocation also included a mistyped supplemental
# DefaultPublishedHandbookTests selector: 1,234 total, rc1 solely for that loader error.
# The corrected supplemental check below passes; real modules were not rerun.

python3 -B -m unittest \
  tests.test_adoption_bootstrap_macos.PlatformPageTests.test_retirement_decision_provides_a_bounded_pinned_inverse \
  tests.test_new_wsl_handbook.NewWslHandbookTests.test_the_committed_outputs_are_current_and_the_handbook_receipt_names_them
# 2 cases, rc0, including the corrected handbook receipt check.

python3 scripts/component_matrix.py --write
python3 scripts/new_host_grand_list.py --write
python3 scripts/evidence_manifest.py --write
python3 scripts/validate.py
# FULL rc0: 70 components, 11,426 hashed files, 4 profiles, 239 receipts.

actionlint .github/workflows/adoption-bootstrap.yml
zizmor --no-online-audits .github/workflows/adoption-bootstrap.yml
# Both rc0; zizmor reports no findings (2 existing suppressions).

gitleaks git --redact --log-opts="$(git merge-base origin/main HEAD)..HEAD" .
# PR range only; rc0, no leaks.

Scratch inverse outcomes and retained failed attempts are recorded in docs/decisions/2026-10-10-retire-macos-ci.md. The initial whole-old-module import found an unrelated removed verdict-review job; bounded extraction resolves it. The initial scratch registry failure was corrected by re-registering own changes relative to main. These checks establish local reversibility and integrity; native macOS execution and acceptance remain unmeasured.

Decision record

docs/decisions/2026-10-10-retire-macos-ci.md, including the superseded decisions, revisit triggers, executable inverse and native macOS limitation.

Host evidence

No host receipt or platform acceptance change is part of this retirement.

Checklist

• Retained third-party actions use full commit pins and version comments.
• Top-level permissions: {} and existing job-local minimum scopes remain.
• Credentials are not printed or committed; no required secret was added.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned files and worktrees were preserved; the owned restoration scratch is removed after verification.