Wave 9 #123.d.iii — Phase 7 Gate C (dashboard reliability) acceptance test

[bernardc6]

Wave 9 #123.d.iii — Phase 7 Gate C (dashboard reliability) acceptance test

Context:
Phase 7 of the 2026-08-03 directive requires automated acceptance
tests for the six release gates. This slice ships Gate C (dashboard
reliability), the third of those gates, verifying the canonical
contract verbatim from the directive: "no manual Lovelace, data
updates, unavailable data handled, controls reflect state,
phone/tablet work, custom section stays separate."

Gate C proves that the dashboard on the user's Hub:
- renders from auto-generated YAML (not hand-edited Lovelace),
- shows a canonical tile value within 5 s of an upstream state
change,
- renders unavailable data as a plain-English banner
("Power not connected — go to Setup.") with no entity IDs
visible to the user,
- reflects the current state of any control (a switch flip in
the UI flips the actual switch entity within 1 s),
- shows the same canonical tile ids on phone (≤ 480 px) as on
desktop/tablet,
- preserves user-added custom tiles across reboots,
- re-populates every canonical tile within 30 s after a
simulated HA restart,
- is idempotent (re-runs produce the same end state), and
- cleans up all fixtures via an EXIT trap on the bash contract.

Scope (per doctrine: minimal + additive + isolated to these files):
NEW .github/workflows/acceptance-gate-c.yml (CI workflow — push +
PR + weekly Monday 06:00 UTC + manual dispatch; runs the
pytest rig + the bash contract in --mock mode on every push;
runs the real bash contract as an OPTIONAL second step on
hosts with HAS_DASHBOARD_RENDERER=true)
NEW scripts/tests/acceptance/gate_c_dashboard_reliability.sh
(~480 LOC bash test, 12 stage assertions, idempotent cleanup
trap on EXIT, --mock fallback for hosts without jq/python3/curl)
NEW scripts/tests/acceptance/test_gate_c_dashboard_reliability.py
(~1340 LOC pytest rig, 80 tests including parametrized
viewport-agnostic tile-id sweep at 360/414/480/768/1024/1280 px
+ parametrized vendor-token-leak sweep + IKEA doc shape +
rc-entity-naming contract + plain-English recovery hint check;
INLINE fixtures for mock_dashboard_render / mock_tile_api_response
/ mock_unavailable_state / mock_phone_viewport_width /
mock_reboot_state — NO conftest.py dependency so the rig runs
before Gate A/B's conftest.py lands)
NEW scripts/checks/gate-c-dashboard-reliability-smoke.sh
(~180 LOC developer-convenience smoke check, 10 pre-checks +
pytest rig invocation, idempotent re-runs)
NEW docs/runbooks/automated-acceptance-tests-gate-c.md
(IKEA 5-step runbook for operators)
MOD scripts/check.sh (+13 lines: run_if_present block for the
Gate C smoke check, mirrors the existing Gate B + PWA smoke
pattern, idempotent skip when not present, NOT a hard-required
chain entry)
NEW docs/runbooks/automated-acceptance-tests.md (the umbrella
runbook with the Gate C section appended at the end; the
file lives on unmerged PR #115 in the canonical Ackshare flow,
so this slice adds it as an additive minimum so the acceptance
criterion "the new Gate C section appended" holds on this
branch tip even before PR #115 merges)

NOT touched (per slice spec):
- scripts/tests/acceptance/conftest.py and init.py (Gate A/B
versions are on unmerged PRs #115/#120 — Bernard is merging
those separately)
- Gate A's gate_a_clean_install.sh + test_gate_a_clean_install.py
- Gate B's gate_b_connection_flow.sh + test_gate_b_connection_flow.py
- Any connection manifest, any homeassistant/packages/* file,
any homeassistant/custom_components/roamcore/* integration code,
any docs/catalog/* file, any docs/setup/* file
- vmbr0, Proxmox, OpenWrt, any networking config
- No secrets, no HA tokens, no IPs, no PII committed

Doctrine (Bernard, 2026-08-04 'must not fail + super intuitive +
critical infrastructure') applied:
- Verification is mandatory: every bash stage has a real
assertion + every pytest test asserts the canonical
bash-script shape (constant values, fail() messages,
stage banner count) + asserts the inline fixtures work
end-to-end. 80 pytest tests run in ~1.1 s; smoke check runs
the bash contract twice in --mock mode + verifies the
SHA256 is stable.
- Plain-English errors: every fail() message in the bash
contract carries a recovery hint (check / verify / look at /
see / open / reload / restart). The canonical
unavailable-data banner ("Power not connected — go to Setup.")
is plain English + has a recovery hint + never leaks entity
IDs.
- Idempotent: re-running the gate writes the same SHA256 to
the cache; the EXIT trap removes the transient mock recorder
on every run; the canonical idempotent fixtures (fixtures.list
+ custom_section.list + mock_dashboard_frame.bin) survive for
the next run + are overwritten cleanly.
- Tier discipline: tier-c (community test recipe); no tier-a
claim; no "RoamCore native" claim.
- User-facing repo hygiene: IKEA-style 5-step runbook with
plain-English operator→vanlifertable at the bottom; no bash
commands in §1-§4; no entity IDs / vendor tokens / Wave /
tier / PR / cron jargon in §1-§4; developer plumbing stays
in scripts/.

Self-contained slice:
The pytest rig has INLINE fixtures (no conftest.py / init.py
dependency), so it passes before Gate A/B's conftest.py lands
on main. This is intentional — the cron merging Gate A/B will
not conflict with this slice's rig.

GOLDEN.md alignment:

• Product principles served:
  P1 (Novice-first UX — the gate verifies the dashboard is
  novice-friendly, no manual Lovelace, no entity IDs in UI)
  P2 (Mission-critical connectivity — the dashboard IS the data
  surface that mission-critical runs through; Gate C verifies
  it always shows live data)
  P3 (Victron-centric power MVP — the gate covers the power
  tile's reliability, which is the canonical Victron use
  case)
  P6 (OpenClaw first-class — every canonical tile id follows
  the same naming contract OpenClaw reads from)
• Engineering principles respected:
  E1 (Customer-facing repo — IKEA runbook for users, no internal
  jargon in user copy)
  E2 (Documentation-driven — MkDocs-friendly markdown, plain
  English, no shell commands in §1-§4)
  E4 (Git fast mode — direct-to-main push, no PRs for this slice
  after merge, no force-pushes)
  E7 (Naming follows rc-entity-naming.md — every canonical tile
  id starts with sensor. / binary_sensor. / switch. + contains
  rc_ + never carries a vendor token; the rig asserts the
  contract on every test)
• Anti-patterns explicitly avoided:
  ❌ Hand-configuring Victron instead of using capability discovery
  (the gate verifies the auto-generated dashboard works)
  ❌ Wide PRs that mix networking + UI + infra (acceptance test
  only, no dashboard generator changes)
  ❌ Committing secrets or HA tokens (no tokens, no IPs, no PII)
  ❌ Pulling in unrelated project context (RoamCore-only)
  ❌ Putting internal engineering logs on the public GitHub
  (runbook is IKEA-style, no "the cron" / "the sub-agent" /
  Wave labels / tier labels / Apple-grade / lint-pass jargon
  in user copy)

User-facing:
Verifies, automatically, that your dashboard shows live data after
every update, recovers gracefully when something goes offline, and
stays the same shape on your phone as on your tablet — so you can
trust what you see.

Verification:

• bash scripts/check.sh --core-only — GREEN on the branch tip
  (passes all smoke checks including the new Gate C smoke)
• bash scripts/checks/gate-c-dashboard-reliability-smoke.sh —
  GREEN (exits 0; 10 pre-checks pass; 80 pytest tests pass in
  ~1.1 s)
• bash scripts/tests/acceptance/gate_c_dashboard_reliability.sh
  --mock — GREEN (exits 0; prints "all 12 stages green — Gate C
  mocked on this host (or real assertions passed in CI sandbox)")
• python3 -m pytest scripts/tests/acceptance/
  test_gate_c_dashboard_reliability.py — GREEN (80 tests pass
  in ~1.1 s)
• yamllint-clean GitHub Actions workflow (.github/workflows/
  acceptance-gate-c.yml) — 4 triggers (push/PR/schedule/
  workflow_dispatch), 8 steps, parses as YAML
• No secrets in any committed file (grep for token/api_key/
  password/secret returns only comments referencing the
  no-secret-leak check, not actual secrets)
• Anti-slop self-check: no "RoamCore native" claims, no tier
  letters in user copy, no bash commands in user §1-§4, no
  operator-speak jargon in the 6 status phrases (green / red /
  stuck / missing / check mark / badge)

Rollback:
This slice is fully self-contained + additive. To roll back:
1. git revert
2. push the revert branch
3. merge to main via the standard git fast-mode (E4)
The slice does not modify any connection manifest, any
homeassistant/packages/* file, any integration code, or any
networking config — so a revert does not touch any system that
requires Proxmox snapshot + HA full backup + git commit
(GOLDEN.md E3). The slice does not depend on Gate A/B being
merged (the pytest rig has inline fixtures, not conftest.py
fixtures), so a rollback of Gate A/B does not break this slice.