The mess
A market-data feed stopped delivering. Nothing failed: the loading job ran on schedule, found no new file, and did nothing. The daily risk report kept publishing, showing the last good day as though it were today. The numbers looked normal, because they were real numbers — just old ones.
Nobody noticed until an auditor compared a report date with the prices behind it.
The method
The fix was not a better dashboard. It was making every stage of the platform assert on the data’s own dates, not on whether a job ran. A feed whose newest record is older than its agreed limit now stops the run, and the report refuses to publish rather than show a stale snapshot.
Each feed got a written freshness rule, an owner, and an alert that reaches a person, not a shared inbox.
The result
A stale feed is now a visible, named failure within hours, not a silent one discovered weeks later. The risk report either shows today’s data or says clearly that it cannot.
What made it stick
The freshness rules are tested, including the uncomfortable cases: an empty file, a file with yesterday’s date, and a feed that delivers twice. When a new feed is added, it cannot go live without its own rule.