Every problem note carries tags on three axes. At most two per axis. Three or more on one axis means the document is trying to be two documents.

target/ — what broke

The only axis allowed to grow. One tag per system actually involved.

target/postgres, target/pgbouncer, target/kubernetes, target/nginx, target/kafka, target/redis, target/aws-rds, target/istio, …

layer/ — where the cause finally was

Not where it hurt. Where it was. Hard cap: 12 values. Adding a thirteenth means one of these is wrong.

layer/network, layer/storage, layer/database, layer/compute, layer/auth, layer/config, layer/dns, layer/observability, layer/capacity

Leave the layer tag off while root_cause is still empty. Guessing here poisons the map.

symptom/ — what shape it had

symptom/outage, symptom/intermittent-failure, symptom/latency, symptom/resource-exhaustion, symptom/data-inconsistency, symptom/silent-failure, symptom/deploy-failure

Off-axis

Exactly one: meta, for the operating documents of this site. Nothing else. Create a new tag only when no existing one covers the case — and for layer/, essentially never.

Related: capture workflow, incident map.