## Implementation Plan

### 1. Modify `NewsService` fetch with retry, audit, and notification

**File:** `devplacepy/services/news/service.py`

- **Import `retry_send`** from `devplacepy/services/openai_gateway/reliability.py`
- **Add config fields** (near existing `ConfigField` definitions):
  - `news_fetch_retries` – type `int`, default `3`, min `0`, max `10`, group `"Reliability"`
  - `news_fetch_retry_backoff_ms` – type `int`, default `5000`, min `1000`, max `60000`, group `"Reliability"`
- **Replace the bare `client.get()` call** (around line 170) with:
  ```python
  response, exc, attempts = await retry_send(
      do_call=lambda: client.get(api_url, timeout=timeout),
      max_retries=self.get_setting("news_fetch_retries"),
      backoff_ms=self.get_setting("news_fetch_retry_backoff_ms"),
      log=lambda msg: self.log(msg),
  )
  ```
- **On failure (all retries exhausted):**
  - Call `audit.record_system("news.service.fetch_failed", ...)` – include `api_url` and `attempts`.
  - Call `_notify_admins` (implement inline or as a helper) using `create_notification` from `devplacepy/utils/notifications.py`:
    ```python
    for admin in get_table("users").find(role="Admin"):
        create_notification(
            admin["uid"],
            "system",
            f"News API unreachable after {attempts} attempts",
            related_uid="",
            target_url="",
        )
    ```
  - Log a final warning.

### 2. Update tests

**File:** `tests/unit/services/news.py`

- Modify `FailingApiClient` to track call count:
  ```python
  class FailingApiClient(FakeClient_news_service):
      def __init__(self, *args, **kwargs):
          super().__init__(*args, **kwargs)
          self.call_count = 0
      async def get(self, url, timeout=None):
          self.call_count += 1
          if url == API_URL:
              raise httpx.HTTPError("api down")
          return FakeResp_news_service(text="")
  ```
- Update `test_run_once_handles_api_failure` to assert that `call_count == news_fetch_retries + 1` (default 4). Use `monkeypatch` to set the new config fields.
- Add a new test `test_run_once_handles_api_failure_zero_retries` that sets `news_fetch_retries=0` and asserts `call_count == 1`.
- Mock `create_notification` (or `get_table`) to verify notification is called exactly once per admin on failure.

### 3. Verify audit event key registration

No change needed – `categories.py` line 27 already registers `"news"` as a prefix, so `news.service.fetch_failed` is valid.

### 4. Ensure consistency with existing patterns

- The `retry_send` function from `reliability.py` uses linear backoff (backoff_ms * attempt). This is acceptable for the stated use case.
- The admin notification pattern mirrors `devplacepy/routers/issues/comment.py` exactly.

---

## Definition of Done

- [ ] **All code changes** as described above are present and committed.
- [ ] **Lint passes:** `pip install -q ruff && ruff check .` returns exit code 0.
- [ ] **Unit tests pass:** `pip install -e '.[dev]' -q && make test-unit` returns exit code 0.
- [ ] **New tests pass** and include:
  - Assertion that `FailingApiClient.call_count == max_retries + 1` on persistent failure.
  - Assertion that `create_notification` is called for each admin on failure (or at least once).
  - Edge case with `news_fetch_retries=0` (single attempt, no retries).
- [ ] **Manual verification** (optional but recommended): run `run_once()` with a mock API that fails; observe three retries, then an audit event and notification in log/database.