Name ops events by object, not job ID
Job IDs feel precise until two teams track the same load under different codes and your freshness chart silently splits.
- naming
- instrumentation
Application analytics for cloud data operations collapses when every squad invents its own namespace for the same pipeline action. Prefer a shared vocabulary: object, action, and context—written once and reused.
Ban temporary ticket codes inside event names. Properties can carry job run IDs, environment, and experiment flags; the event name should still make sense after a reorg.
Store the dictionary where operators already look—repository docs or a shared sheet with owners—not only in a slide from last quarter’s offsite.
In Pipeline Visibility Audits from Tuen Mun we routinely find dozens of near-duplicates of the same completion signal. Collapsing them usually unlocks more clarity than adding another chart.