The local FastAPI operations dashboard — a workflow runner with per-feature scope, persisted run history, workflow chaining, and live SSE log streaming

Failure modes

Symptom	Cause	Fix	Severity
`RuntimeWarning: coroutine 'RunnerService.start' was never awaited`	`start()` called without `await`	It's async — `await` it / `asyncio.run`	high
Workflows won't run from the UI	`Config.allow_run` is `False` (e.g. `--read-only`)	Start without `--read-only`	high
`RunnerBusyError` on start	A run is already active (one at a time)	Wait for / inspect `runner.current`	medium
`TypeError` calling `config.runs_dir()`	`runs_dir` is a property	Drop the `()`	medium
Cost panel shows an error	`fetch_summary` returned a `CostFetchError` (no admin key, network)	Inspect the error `kind`/`message`; cost is an adjacent surface	low

RunnerService.start and Run.subscribe are async. They are the two awaitable surfaces; everything else on the runner is sync.
allow_run is a real gate. Off by default in Config; the CLI turns it on unless --read-only.
Scope confusion. Cost/telemetry/help shown on the dashboard are adjacent surfaces — owned by other features, only read here.

Confirm the app builds: create_app(build_config(Path("."))).
Confirm runs are allowed: config.allow_run (and no --read-only).
For a run that won't start: check runner.current / RunnerBusyError.
For async warnings: await start() / iterate subscribe().
For a data panel: that data is owned by its feature (telemetry / help-system / cost).