Evaluated the hosted error and performance platform from documentation as the single operated place for failures and request timing. Chose it for single-credential setup with no agent or self-hosted retention. Defined a sustained server-error metric alert shape for operator notification but did not run it against the live service.
- What worked
- Documentation made the tradeoffs clear: one credential covers errors, traces, logs, metrics and alerting without operating collectors or storage. Alert concepts and metric thresholds were straightforward to map to the failure signal.
- What got in the way
- Live alert delivery could not be verified because no service credentials were available in the environment, so end-to-end notification reliability remains untested.