Authored by Tharun Mathew. Featured in Platform Engineering.
Every container is healthy, every pipeline run is green, and the number in the revenue table is still wrong. Nobody gets paged, because nobody owns it.
This article argues that platform engineering drew its ownership lines around the runtime. That works well for microservices, where the thing you own, the thing you measure and the thing that fails are all the same. Data breaks that pattern. A service fails loudly: the process dies, the probe fails and the deployment rolls back. A dataset fails quietly, carrying a currency mismatch or a join that dropped rows while every health check still passes. The result is that teams can name who owns the cluster, the pipeline and the on-call rota, but not who is accountable for whether the numbers are right.
Tharun sets out three warning signs: nobody can say what a dataset promises, schema changes go straight to production without anyone downstream being consulted, and datasets nobody reads keep running up costs because no one can safely delete them. The usual fixes, a data catalogue or a central data team, do not close the gap. Instead, the piece defines what real ownership needs: a named owner on the rota, a stated interface, known consumers and a deprecation path with a real end date. Rather than writing another policy, it makes the case for treating data contracts as declarative infrastructure, with CI/CD rejecting any pipeline change that lacks a schema specification, an alert routing key and a retention manifest.
The impact was measurable. The time taken to find an accountable owner for a dataset fell from around three days to under an hour, and roughly a third of catalogued datasets turned out to have no living consumer. Those were retired within two quarters, removing spend nobody could see.
It closes with a question every platform team should be able to answer: if a figure in one of your datasets is wrong today, how long would it take someone to notice, and whose name is attached to it?
Read the full article here: https://platformengineering.com/features/platform-teams-own-the-pipeline-nobody-owns-the-data/
