Monitoring

• alerts.yml: Prometheus alert rules: no clock ticks, versions in default partitions, feed consumer lag, dead sink batches, failed merges and thaws, an unsealed ledger, branches over quota, conflicts left open.
• grafana-dashboard.json: a Grafana dashboard over the same metrics (import it and pick the Prometheus data source).

The metrics are gnomon.metrics(): gnomon serve prints them on /metrics, or point postgres_exporter at the function (gnomon.metrics_text() returns the text form). They read catalogs and statistics only. Per-branch series (gnomon_branch_behind_main) are off unless asked for (gnomon.metrics(per_branch => true), /metrics?per_branch=1) and capped, to bound cardinality. Not exposed yet: the result of the last ledger verification and the C module's counters (they are per session, not in shared memory).