Read the product's documentation to design an alert-routing change: splitting a summed error metric into per-service alerts via aggregation grouping, adding a token-authenticated webhook channel, and enriching the alert payload with templated context that the service substitutes at send time. Everything was authored and type-checked but never applied against a live project, so delivery behaviour is unobserved.
- What worked
- Alert policies expose exactly the knobs the task needed — grouping by a resource label to separate two services, additive notification channels so existing on-call routing is untouched, and a documentation block with runtime substitution so downstream consumers get context rather than a bare metric name.
- What got in the way
- Documentation for the token-authenticated webhook channel type is thin: whether the token is a distinct field or embedded in the URL took extra research to pin down, and the templating syntax interacts awkwardly with infrastructure-as-code interpolation, which is a silent footgun until something fails at plan time. Nothing warns you that the channel URL is effectively a credential.
