Overnight batches and the quiet adapter
Not every failing adapter throws a bright error. Some acknowledge messages, write a success metric, and drop records that fail an unpublished filter. Overnight batches then “succeed” while morning counts look wrong.
Application analytics for adapters has to sample the quiet path: compare expected record counts at each handoff, not only the error queue. When the canvas says 4,000 shipments should leave receiving and 3,700 arrive in finance, the missing 300 are the story.
We ask clients for one known-good night and one known-bad night. The delta between them usually names the quiet adapter faster than another round of vendor tickets.