Your service catalogue says every service has an owner. Investigating an incident, you find the named owner left the company eight months ago. How do you prevent this class of problem?

What the interviewer is testing

Whether you understand that declared metadata rots and derived metadata does not — the single most important property of a catalogue.

The root cause

The field was populated by hand and nothing verifies it. Reorganisations happen far more often than services are retired, so ownership is the fastest-decaying field in any catalogue and the one most needed during an incident.

Worse than absent, it is misleading: people trust it and page someone who does not exist.

The fixes

Own by group, not person. The owner is a team identity — a group in the identity provider, a rota, a distribution list — that survives individuals leaving. Individual names belong in the rota, which has its own lifecycle.

Derive rather than declare. Take ownership from repository metadata that engineers maintain because it affects their daily work, not from a catalogue field they filled in once.

Verify continuously. A scheduled check that every service's owning group exists, is non-empty, and has an active on-call rota. Failures raise a ticket against the parent organisational unit, which is the only party who can resolve it.

Make it a deployment gate for new services: no owner, no production. Cheap at creation, impossible to retrofit.

The wider principle

A catalogue's value is a direct function of how much of it is derived — schemas from the platform, dependencies from build manifests and traces, versions from the deployment system, health from monitoring, popularity from access logs. Declared fields should be the minimum irreducible set, and each one needs a verification mechanism or it will be wrong within a year.

What a strong answer adds

Attaching the orphan check to a reorganisation trigger: when a group is dissolved in the identity provider, every service owned by it is flagged immediately rather than discovered eight months later during an incident.

Common weak answers

An annual review campaign, which is accurate for a week. Adding more required fields, which increases the rot surface.