Case Studies
What we did, how we measured it and what came out. Every number comes with its sample and date.
Grafana alerting audit: 16 of 209 alert rules could never fire
We asked not whether an alert is firing, but whether it can fire at all. For each of 209 rules we counted the series its query returns through the data source the rule itself uses, and checked the base metric without thresholds. Sixteen rules were blind by mistake; all are closed, and since 22 September a daily watchdog tells 'zero by design' from an orphaned rule.
Data completeness: a pipeline that reports its own shortfall
A market analytics system collects open data on procurement and job listings. An audit showed that a partial collection never turned into a failure anywhere, and the card collector did not even have a denominator. We added a denominator, a second pass and a 'fully collected' verdict, and made the checks strict only where measurement showed zero shortfall.