cc-ci/machine-docs/STATUS-2.md

# STATUS — Phase 2 (per-recipe test authoring)

**Phase plan (SSOT):** `/srv/cc-ci/cc-ci-plan/plan-phase2-recipe-tests.md`
**Loop state for THIS phase:** STATUS-2 / BACKLOG-2 / REVIEW-2 / JOURNAL-2 (DECISIONS.md shared).
Phase 1/1b/1c/1d/1e STATUS/BACKLOG/REVIEW files are HISTORY (all DONE) — not this phase's state.

## Phase
Phase 2 authors **per-recipe test content** on top of the corrected Phase 1/1d/1e shared harness.
Per the plan, for every maintained Co-op Cloud recipe (§5 target set), the cc-ci `tests/<recipe>/`
tree must carry:
- Phase-1d/1e **lifecycle overlays** (assertion-only, additive) — `test_install.py`, `test_upgrade.py`,
  `test_backup.py`, `test_restore.py` + `ops.py` pre-op seeds.
- **Parity-ported** tests from `references/recipe-maintainer/recipe-info/<recipe>/tests/*.py`,
  one-to-one (P2), with a `PARITY.md` mapping table.
- **≥2 NEW recipe-specific functional tests** (P3) — characteristic behavior, not just `status==200`.
- **Real backup data-integrity** (P4): seed → backup → mutate → restore → assert seeded data survived.
- **Dependency resolution** (P5): recipes that need other apps (SSO providers, DBs) deploy them in-run.
- **Playwright** (P6) where the app's core UX is a UI flow.
- **Docs** (P8): `docs/enroll-recipe.md` updated with the per-recipe test contract + worked example.

## Definition of Done (Phase 2) — P1–P8, each Adversary cold-verified in REVIEW-2
- [ ] **P1 — Coverage.** Every recipe in §5 target set has a `tests/<recipe>/` suite enrolled and a
      full green `!testme` run (install + upgrade + backup-restore).
- [ ] **P2 — Parity port.** Every `recipe-info/<recipe>/tests/*.py` has a comparable cc-ci test;
      `tests/<recipe>/PARITY.md` records the mapping; non-ports documented in DECISIONS.md.
- [ ] **P3 — Recipe-specific depth.** Each recipe has **≥2 new functional tests** beyond parity
      (characteristic behavior, real assertions on app state/responses).
- [ ] **P4 — Backup data-integrity is real.** Seed → backup → mutate → restore → assert seeded data
      survived (recipe-aware, not health-only). Pattern already proven in Phase 1e on custom-html.
- [ ] **P5 — Dependencies handled.** Recipes with deps declare them; harness deploys deps within the
      run (respecting `MAX_TESTS`); SSO setup runs automatically.
- [ ] **P6 — Browser flows where they matter (D3).** UI-centric recipes have a Playwright test of the
      core flow (login, create-an-object, etc.).
- [ ] **P7 — No weakened tests, no corners cut.** Every assertion is real; nothing skip/xfail'd,
      mocked, or health-only stand-in. Any "untestable" claim is a true env-level blocker with
      Adversary sign-off.
- [ ] **P8 — Docs.** `docs/enroll-recipe.md` updated with the per-recipe test contract (§4.1) and a
      worked example; a new engineer can add a recipe's full suite from the docs.

## Milestones (plan §6)
- **Q0** — Harness additions (HTTP/convergence, OIDC-flow, dep resolver, backup data-integrity, TTY
  abra). Reference recipe (custom-html) uses them for full parity+specific suite, green via `!testme`.
- **Q1** — Pattern proof (custom-html + n8n): full parity + ≥2 specific + real backup data-integrity.
- **Q2** — SSO providers (keycloak + authentik); reusable SSO-setup/OIDC-flow harness e2e.
- **Q3** — SSO-dependent suite (lasuite-docs, lasuite-drive, lasuite-meet, cryptpad, immich); deps
  auto-deployed, SSO setup automated, parity + specific.
- **Q4** — Remaining recipes (matrix-synapse, mumble, bluesky-pds, ghost, mattermost-lts, discourse,
  plausible, uptime-kuma, mailu, drone).
- **Q5** — Completeness + docs; flip `## DONE`.

## In flight
**Q3 + Q4 — recipe enrollment sprint.** After capacity unblock + Adversary checkpoint, landed:
- Q3.1 lasuite-docs partial (parity + 2 specific + Q2.4 test_oidc_with_keycloak); deeper OIDC
  ports deferred in DEFERRED.md.
- Q3.4 cryptpad partial (parity + 2 specific); create-pad deferred F2-9 conditional (must lift
  before Phase-2 DONE).
- Q4.1 matrix-synapse FULL (parity-aligned + 3 specific incl. §4.3 register-and-message).
- Q4.3 bluesky-pds FULL (4 functional incl. §4.3 account+post round-trip via goat CLI; F2-8 closed).
- Q4.4 ghost FULL (parity + 3 specific; create-post deferred in DEFERRED.md).
- Q4.8 uptime-kuma FULL (parity + 2 specific; create-monitor deferred in DEFERRED.md).

Harness change: `lifecycle.deploy_app` + `run_recipe_ci.py` + `deps.py` now thread
`recipe_meta.DEPLOY_TIMEOUT` into `abra.deploy(timeout=...)` so heavy-recipe Python subprocess
timeout matches the recipe's internal TIMEOUT.

DEFERRED.md (machine-docs/) — new orchestrator-canonical deferral registry; 9 entries open.

Remaining substantial: Q3.2 lasuite-drive (needs mirror), Q3.3 lasuite-meet (mirrored), Q3.5
immich (needs mirror), Q4.2/Q4.5-7/Q4.9-10 (mostly need mirror). The mirror-and-enroll path is
established (recipe-create-pr skill); pausing this sprint for Adversary cold-verify.

## Adversary findings — Builder response

**F2-11 — FIXED, awaiting Adversary re-verify** (commit: `git log --oneline | grep 'F2-11'`).
SSO-dep "deps-not-ready"
SKIP no longer yields a GREEN `!testme`.
- **WHAT:** when a recipe declares `DEPS` and `setup_custom_tests` fails (deps not ready) so its
  `@requires_deps` (SSO/OIDC) tests SKIP, the run now reports **FAIL** (`overall=1`), not green —
  while generic-tier failure-isolation is preserved (install/upgrade/backup/restore results stand).
- **WHERE (code):**
  - `tests/conftest.py::pytest_collection_modifyitems` — now counts the requires_deps tests it skips
    and appends the count to `$CCCI_DEPS_SKIP_REPORT`.
  - `runner/run_recipe_ci.py` — sets `CCCI_DEPS_SKIP_REPORT` (run-scoped temp, near `depsfile`);
    after teardown sums the count into `requires_deps_skipped`; RUN SUMMARY annotates the custom tier
    (`custom: pass (N requires_deps SKIPPED ... SSO UNVERIFIED)`); new pure predicate
    `sso_dep_unverified(declared, deps_ready, requires_deps_skipped)` flips `overall=1`.
  - `tests/unit/test_f211_sso_skip.py` — 7 new unit tests.
- **HOW to verify (both deploy-free, rate-limit-independent):**
  1. `ssh cc-ci 'cd /root/cc-ci && cc-ci-run -m pytest tests/unit -q'` → **EXPECTED: 35 passed**
     (28 prior + 7 F2-11).
  2. Cold real-test signal proof:
     `ssh cc-ci 'cd /root/cc-ci && rm -f /tmp/f211-skip.txt && CCCI_DEPS_READY=0 \
       CCCI_DEPS_NOT_READY_REASON=boom CCCI_DEPS_SKIP_REPORT=/tmp/f211-skip.txt \
       cc-ci-run -m pytest tests/lasuite-docs/functional/test_oidc_with_keycloak.py -rs; \
       cat /tmp/f211-skip.txt'`
     → **EXPECTED:** `1 skipped`, pytest exit 0 (the hazard), and `/tmp/f211-skip.txt` == `1`. Since
     lasuite-docs declares `DEPS=["keycloak"]`, the orchestrator computes
     `sso_dep_unverified(["keycloak"], False, 1)=True` → `overall=1`.
- **NOT verified by a live run yet:** full e2e (real deploy with forced setup_custom_tests failure →
  observe `overall=1`) is deferred until the Docker Hub rate limit (## Blocked) lifts. The two proofs
  above cover the predicate, the conftest signal on real files, and the count flow; only the
  straight-line read→sum→predicate→overall wiring is unexercised by a live deploy.

## Gate
**Gate: Q2 — Adversary PASS @2026-05-28** (REVIEW-2 `## Q2 — PASS @2026-05-28 (re-verify after
F2-5 fix + F2-6 collateral resolution)`; cold e2e on `/root/adv-verify` HEAD `874bfbb`:
deploy-count=2, all 5 assertions PASS, DEPS teardown clean, post-run docker stack/volume/secret
with 'keyc|lasuite' filter all empty; NO VETO). F2-5 + F2-6 CLOSED; F2-7 stands as open scope
(authentik backend in harness.sso when Q2.2 enrolls). Builder may advance to Q3 — already in
flight (Q3.1 partial @ `874bfbb`, Q5.1 docs @ `b2151af`).

Acceptance per plan §6 Q2: "a dependent recipe deploys its provider + runs an OIDC login test
in one run." Proven cold:

**Objective evidence pointers (Q2):**
- **Q2.1 keycloak parity + 2 NEW specific tests** — commit `d5f5e86`:
  - `tests/keycloak/functional/test_health_check.py` — parity port.
  - `tests/keycloak/functional/test_password_grant_token.py` — password grant, JWT decoded, claims
    (iss/azp/typ/exp/iat) validated.
  - `tests/keycloak/functional/test_create_client_and_use.py` — admin-API client CRUD +
    client_credentials grant + JWT azp/iss validation + idempotent cleanup.
  - `oidc_integration.py` parity deferred to Q3 (cross-recipe; see PARITY.md note).
  - Bumped DEPLOY_TIMEOUT + HTTP_TIMEOUT to 900s.
  - Cold e2e (log `/root/ccci-q2-keycloak-r3.log`): all 5 stages PASS, deploy-count=1,
    `head_ref=666649a6 == chaos-version=666649a6`, version `10.7.0+26.6.1 → 10.7.1+26.6.2`.

- **Q2.3 dep resolver + SSO-setup harness primitives** — commit `4d6b040`:
  - `runner/harness/deps.py` — declared_deps + dep_domain + deploy_deps + teardown_deps + JSON
    run state. Subsumes Q0.4 (dep resolver).
  - `runner/harness/sso.py` — setup_keycloak_realm + oidc_password_grant +
    assert_discovery_endpoint. Reusable by every SSO-dependent recipe (Q3 will exercise).
  - `runner/run_recipe_ci.py` — wired in dep deploy BEFORE recipe-under-test, dep teardown
    AFTER in finally (reverse order). DG4.1 expected count = 1 + len(deps).
  - `tests/conftest.py` — `deps_apps` fixture exposes dep domains to dependent tests.
  - 7 new unit tests in `tests/unit/test_deps.py`; **28/28 unit tests PASS** cold.

- **F2-5 fix — dep teardown verify=True** — commit `c6e94af`, log `/root/ccci-f25-verify.log`:
  - `runner/harness/deps.py::teardown_deps` now uses `lifecycle.teardown_app(..., verify=True)`
    so residuals raise `TeardownError`. Errors are logged per-dep but we continue to other deps;
    a combined `TeardownError` is raised after all attempts.
  - `runner/run_recipe_ci.py` catches the dep `TeardownError` in finally, surfaces via
    `dep_teardown_error` in the run summary + non-zero exit code.
  - Cold-verified: lasuite-docs+keycloak dep e2e PASSED clean (3 custom + 2 lifecycle install =
    5 PASS); post-run cc-ci state has NO leftover keycloak (`docker stack ls | grep keyc` →
    empty; `docker volume ls | grep keyc` → empty; `docker secret ls | grep keyc` → empty).
  - deploy-count=2, expected 2.

- **Q2.4 acceptance (the gate)** — commit `9e88741`, log `/root/ccci-q24-lasuite-keycloak.log`:
  - `tests/lasuite-docs/recipe_meta.py` declares `DEPS = ["keycloak"]`.
  - `tests/lasuite-docs/functional/test_oidc_with_keycloak.py`:
    - Asserts `deps_apps["keycloak"]` is the per-run dep domain.
    - Calls `harness.sso.setup_keycloak_realm` → realm/client/user.
    - GETs OIDC discovery; asserts `issuer == https://<kc>/realms/lasuite-docs`.
    - Performs password grant → JWT; asserts iss/azp/typ/exp claims.
  - Cold-run output:
    ```
    ===== DEPS: ['keycloak'] =====
      dep: deploying keycloak -> keyc-c12afe.ci.commoninternet.net
      dep: keycloak ready @ keyc-c12afe.ci.commoninternet.net
    ===== TIER: install =====  2 PASS (generic + cc-ci overlay)
    ===== TIER: custom =====   1 PASS (test_oidc_password_grant_against_dep_keycloak)
    ===== DEPS teardown =====
    ===== RUN SUMMARY =====
    deploy-count = 2 (expect 2)
    ```

- **F2-3 systemic fix** — commit `47f7cb4`: `runner/harness/browser.py::goto_with_retry`
  centralizes the F2-3 try/except PlaywrightError pattern; applied to **all** install overlays
  using page.goto (custom-html, n8n, keycloak, cryptpad, lasuite-docs) + the custom-html
  playwright/test_browser_smoke. Cold e2e (custom-html, log `/root/ccci-q2-customhtml-r2.log`):
  all 5 stages PASS, deploy-count=1, HC1 non-vacuous.

**Reference command for Adversary (cold, on cc-ci):**
```
ssh cc-ci 'cd /root/<your-clone> && \
  cc-ci-run -m pytest tests/unit -v && \
  RECIPE=keycloak cc-ci-run runner/run_recipe_ci.py && \
  RECIPE=lasuite-docs STAGES=install,custom cc-ci-run runner/run_recipe_ci.py'
```

---

**Gate: Q1 — Adversary PASS @2026-05-28** (REVIEW-2 `## Q1 — PASS @2026-05-28 (re-verify after
F2-3 + F2-4 fixes)`; cold e2e on `/root/adv-verify` HEAD `fc89552` → all 5 stages PASS,
deploy-count=1, HC1 non-vacuous; F2-3 + F2-4 CLOSED; NO VETO). Builder may advance to Q2.

**Objective evidence pointers (Q1):**
- **custom-html (Q1.1)** — already cold-verified in Q0 PASS. Same evidence stands: full e2e green,
  HC1 non-vacuous, deploy-count=1; PARITY.md + functional/ + playwright/ in place.
- **n8n (Q1.2)** — full e2e on cc-ci (log `/root/ccci-q1-n8n-r3.log`):
  - **HC1 PR-head proof:** `head_ref=63dd3e0f == chaos-version=63dd3e0f`, version
    `3.1.0+2.9.4 → 3.2.0+2.20.6`.
  - **Deploy-count = 1** (DG4.1 holds).
  - **Lifecycle tier results (generic + cc-ci overlay both PASS at each stage):**
    - install: generic `test_serving` PASS + cc-ci `test_serving_and_editor` PASS (the robust
      Playwright poll handles n8n's /healthz-200-before-/-route-registered window).
    - upgrade: generic `test_upgrade_reconverges` PASS + cc-ci `test_upgrade_preserves_data`
      PASS (marker `upgrade-survives` written into /home/node/.n8n by `ops.pre_upgrade` survived
      the chaos redeploy of PR-head).
    - backup: generic `test_backup_artifact` PASS + cc-ci `test_backup_captures_state` PASS
      (marker `original` from `ops.pre_backup` captured by `abra app backup create`).
    - restore: generic `test_restore_healthy` PASS + cc-ci `test_restore_returns_state` PASS
      (marker mutated to `mutated` by `ops.pre_restore`, restored to `original` — real backup
      data-integrity).
  - **Custom tier results (4 PASS — log `/root/ccci-q1-n8n-r4.log` post-F2-4/F2-3 fix):**
    - `tests/n8n/functional/test_health_check.py::test_n8n_returns_200` — parity port (HTTP 200
      from `/`), with `SOURCE: recipe-info/n8n/tests/health_check.py` comment.
    - `tests/n8n/functional/test_workflow_roundtrip.py::test_workflow_create_and_read_back` —
      **plan §4.3 prescribed create+read-back**: owner setup → POST /rest/workflows → GET
      /rest/workflows/<id>; assert id/name/nodes round-trip. (F2-4 fix.)
    - `tests/n8n/functional/test_rest_settings.py::test_rest_settings_returns_json_with_known_keys`
      — polls `/rest/settings` until content-type is `application/json` (rejecting the
      "n8n is starting up" placeholder HTML), then asserts known public-settings keys
      (`userManagement` / `defaultLocale` / `authCookie`) in the `data` envelope.
    - `tests/n8n/functional/test_login_state.py::test_login_endpoint_returns_json` — polls
      `/rest/login` until content-type is `application/json`, proves auth subsystem initialized.
  - **PARITY.md complete:** `tests/n8n/PARITY.md` — parity row for `health_check.py`, rationale
    for the 2 recipe-specific tests, data-integrity + playwright sections.
- **Q1 has no Adversary findings yet.** No tests skipped/weakened; rejecting-the-placeholder
  pattern in the new functional tests is non-vacuous (a stuck-booting n8n that only serves the
  placeholder fails the test).

**Reference command for Adversary (cold, on cc-ci):**
```
ssh cc-ci 'cd /root/<your-clone> && RECIPE=n8n cc-ci-run runner/run_recipe_ci.py'
```

---

**Gate: Q0 — Adversary PASS @2026-05-28** (REVIEW-2 `## Q0 — PASS @2026-05-28`; cold re-verify on
`/root/adv-verify` HEAD `0b834e9` → 21 unit PASS + e2e PASS; NO VETO). F2-1 closed; F2-2 (scope
observation) acknowledged.

**Prior Q0 claim detail (commit `5741e88` — F2-1 fix landed on top of the original Q0 changeset).** Acceptance evidence (per plan §6 Q0): a reference recipe
(custom-html) uses the new harness additions for a full parity + specific suite, green via the
existing run path. F2-1 (test_custom_tests_repo_local_gated stale assertion) closed by Builder; cold
re-run on cc-ci → **21/21 PASS** including the previously-failing test. F2-2 (scope observation:
OIDC-flow + dep resolver not in Q0) acknowledged — those primitives implement when Q2/Q3 consume
them; BACKLOG-2 Q0.4 remains open and explicitly deferred.

**Objective evidence pointers (Q0):**
- **Harness additions landed**
  - `runner/harness/http.py` — canonical Phase-2 recipe-test HTTP API (vendored from
    `references/recipe-maintainer/utils/tests/helpers.py`): `http_get`, `http_post`, `http_request`,
    `retry_http_get`, `retry_http_post`, `wait_for_http`, `assert_converges`. JSON + form bodies,
    transport-failure → status=0.
  - `runner/harness/discovery.custom_tests` recurses into `tests/<recipe>/functional/` and
    `tests/<recipe>/playwright/` (Phase 2 §4.1 layout) while excluding lifecycle `test_<op>.py`
    names; HC2 repo-local gate continues to apply.
  - TTY abra wrapper already present in `runner/harness/abra.py::_run_pty` (Phase 1d) — reused.
- **Unit-test proof (deterministic, cc-ci; post-F2-1 fix commit `5741e88`)**
  - `cc-ci-run -m pytest tests/unit -v` → **21 passed in 5.38s** (the previously-failing
    `test_custom_tests_repo_local_gated` now passes; synthetic-recipe + monkeypatch fixture):
    - 8× pre-existing `tests/unit/test_discovery.py` (overlay + HC2 gate, regressed).
    - 2× new `tests/unit/test_discovery_phase2.py` (functional/+playwright/ recursion + HC2
      gate still applies to subdirs).
    - 11× new `tests/unit/test_http.py` (in-process http.server fixture — JSON parsing,
      4xx-with-body, non-JSON body, transport-failure=0, headers, JSON+form POST, retry
      convergence, retry timeout, wait_for_http, assert_converges return value).
- **End-to-end proof (custom-html on cc-ci, the reference recipe)**
  - `RECIPE=custom-html cc-ci-run runner/run_recipe_ci.py` (log `/root/ccci-q0-customhtml-full.log`):
    - install/upgrade/backup/restore/custom **all PASS**, deploy-count=1.
    - HC1 PR-head proof: `head_ref=8a026066 == chaos-version=8a026066`, version `1.10.0→1.11.0`.
    - 5 lifecycle assertions (generic + cc-ci overlay across 4 ops) + 4 custom-stage assertions
      (3 functional + 1 playwright). Reference command for Adversary cold re-run:
      `RECIPE=custom-html cc-ci-run runner/run_recipe_ci.py`.
- **Per-recipe contract artifact landed**
  - `tests/custom-html/PARITY.md` — parity row for `health_check.py`, rationale for the 2
    recipe-specific tests + the data-integrity + playwright sections.
  - `tests/custom-html/functional/{test_health_check.py,test_content_roundtrip.py,test_content_type_header.py}` — parity port + 2 NEW recipe-specific tests; each parity file carries the
    `SOURCE: recipe-info/custom-html/tests/<file>` comment for audit.
  - `tests/custom-html/playwright/test_browser_smoke.py` — Phase-2 P6 home.

**Reference command for Adversary (cold, on cc-ci):**
```
ssh cc-ci 'cd /root/cc-ci && cc-ci-run -m pytest tests/unit -v && RECIPE=custom-html cc-ci-run runner/run_recipe_ci.py'
```

## Blocked
**(none) — the Docker Hub rate-limit block is RESOLVED @2026-05-28 ~22:10Z. Awaiting Adversary
re-verify of the 3 conditions (immediate relief already confirmed by Adversary in REVIEW-2).**

**Docker Hub rate-limit fix — DONE (registry-creds finding, plan §1.5), all 3 conditions met.**
Operator provided a read-only PAT (`DOCKERHUB_USERNAME=nptest2` + `DOCKERHUB_TOKEN` in `.testenv`).
Wired declaratively; verify commands + expected outcomes for the Adversary:
1. **Authenticated 200-limit from account source** (Adversary already CONFIRMED in REVIEW-2). Re-check:
   `ssh cc-ci` → `docker info | grep Username` = `nptest2`; an authenticated manifest HEAD shows
   `ratelimit-limit: 200;w=21600` and `docker-ratelimit-source: b662dd8b-…` (account hash, NOT IP
   `68.14.43.142`).
2. **Swarm SERVICE-task pulls authenticate** — PROVEN with an **uncached** image:
   `ssh cc-ci 'cd /root/cc-ci && RECIPE=n8n STAGES=install cc-ci-run runner/run_recipe_ci.py'`
   → EXPECTED: `install: pass`, deploy-count=1, NO `toomanyrequests`; the swarm task pulls
   `n8nio/n8n:2.20.6` to 1/1. During the run the **account** counter decrements (197→196 resolution
   →195 agent layer pull, source = account hash) — the agent pull is billed to the account, not the
   anon IP. (n8n images were uncached, so this is a real fresh-pull test, not a cached false-pass.)
   Conclusion: abra `docker stack deploy` propagates the cred on this single-node swarm; no
   `--with-registry-auth` flag or pre-pull needed.
3. **Declarative persistence across a 1c rebuild** — PAT sops-encrypted (`secrets/secrets.yaml` key
   `dockerhub_auth` = base64("nptest2:PAT"), submodule `cdd5e0a`); `nix/modules/secrets.nix` adds
   `sops.secrets.dockerhub_auth` + `sops.templates."docker-config.json"` → renders
   `/root/.docker/config.json` (0600 root) at activation. Verify: after `nixos-rebuild switch`,
   `ls -l /root/.docker/config.json` → symlink to `/run/secrets/rendered/docker-config.json`; the
   activation log shows `adding rendered secret: docker-config.json`. Recorded in DECISIONS.md
   ("Docker Hub auth: declarative config.json via sops").

**Bonus unblocked:** Q3.2 lasuite-drive base deploy now CONVERGES (all 12 services incl.
onlyoffice+collabora) — `RECIPE=lasuite-drive STAGES=install` → `install: pass`, deploy-count=1
(commit before this; the rate limit was the only blocker). Q3.2 specifics (OIDC/WOPI/upload) are next.

**Earlier Gitea outage (RESOLVED @~21:08Z).** git.autonomic.zone returned blanket `404` for ~1.5h
(backend down; same from my sandbox AND cc-ci). Reconciled: pulled + pushed queued commits. The 3
watchdog pings during the outage were phantoms (Adversary's failed push retries); nothing lost.

**Prior bootstrap state:** access re-verified @2026-05-28: `ssh cc-ci` ok (root, NixOS 24.11), Gitea
API HTTP 200, wildcard DNS resolves to gateway 143.244.213.108.

## Carryover from Phase 1e (not blockers for Phase 2)
- **F1e-2** [adversary] — concurrent same-recipe `abra recipe fetch` race in
  `runner/run_recipe_ci.py::fetch_recipe`. Pre-existing in Phase 1d; not a 1e regression. Drone
  caps `MAX_TESTS=1` today, so practical impact bounded. Tracked for Phase-2 breadth-ramp if
  concurrent recipe runs become routine.