mind: file the three CTI CI follow-ups left open by Phase 5

[Jammy2211]

Three prompts for the follow-ups surfaced by the arcticpy / wiki-currency / Phase-5-smoke work shipped 2026-08-24 and deliberately not taken then, so those tasks stayed scoped. Each carries the evidence already gathered, so a fresh session doesn't re-derive it.

draft/maintenance/ci/autocti_workspace_navigator_check.md — priority: high

Heart requires ["Smoke Tests", "Navigator Check"] for the workspaces group; autocti_workspace has only the former.

This is not cosmetic. Verified against heart/checks/ci_status.py's real rollup():

Smoke green, Navigator MISSING -> {'conclusion': '',        'status': 'in_progress'}
both green                     -> {'conclusion': 'success', 'status': 'completed'}
smoke red, Navigator MISSING   -> {'conclusion': 'failure', 'status': 'completed'}

A required workflow with no runs never satisfies all_green, so the repo rolls up permanently in_progress — it can go red but can never go green, and the readiness gate can never see it as CI-clean. Adding Smoke Tests in #28 was necessary but not sufficient.

The prompt also notes that Navigator Check is not a Heart reusable — it lives per-workspace — so the work starts by reading a sibling that has one, not by inventing a check from the name.

draft/maintenance/ci/wiki_currency_check_version_gate.md

The baseline regenerated yesterday will rot again. --check-version gates on a hash of the entire public surface of autoarray and autofit; the red it just produced was 14 symbols, of which not one is cited by the assistant. The workflow installs from the libraries' main clones, so the clock is every export-touching merge into either.

Two options written up — gate on removals only, or accept that --scope all subsumes it as a gate — with the observation that even removals produce false reds here (both removed symbols were uncited), so removals of cited symbols may be the real line. The task is to choose deliberately.

Records that pinning to a released stack was already considered and rejected: autocti's PyPI release is pre-resurrection, so that would grade today's docs vacuously green.

draft/maintenance/ci/heart_smoke_table_autocti.md

config/repos.yaml's smoke: block has no autocti entries and import_names has no PyAutoCTI, so heart/smoke.py cannot prepare or run either CTI smoke suite locally — a blind spot exactly where two suites just landed.

Not a two-line config edit: a composite action can't be called from smoke.py, so the arcticpy recipe needs a local equivalent without re-creating the second copy PyAutoHeart#170 just removed. Three options written up, plus the verified recipe and the two traps (--no-deps suppresses scipy/matplotlib; arcticpy has no __version__).

Sizing note

Declared Difficulty: medium on all three. The sizing faculty derives too-large/too-large/large — but it scores on prompt length (107 lines / 627 words), and these are long because they carry measured evidence, not because the work is large. Flagging the disagreement rather than hiding it.

lifecycle check, index --check OK; dashboard regenerated. (dates reports 2 undated prompts — both pre-existing, neither mine.)

---

Generated by Claude Code