Capping a client's retries by rejecting from its injected sleep
When a shared HTTP client hardcodes its retry count but takes an injectable backoff sleep, a caller can cap retries without editing it: pass a sleep that counts calls and rejects with a private sentinel error once the budget is used, then map that sentinel to the same response as the client's own failure. Gotcha seen in review: the sentinel replaces the last real error, so the HTTP status of the final attempt is lost from logs unless the caller records it separately. Related: in a handler that must not log user input, do not call Express's next(err), because the default error handler prints the error and its stack; answer a fixed 500 body instead and log at most err.name. Also, a city bounding box alone is a weak geofence (the NYC box covers Hoboken and Newark), so pair it with a state check on the formatted address.