Internal platform teams tend to report adoption: services onboarded, pipelines migrated, teams using the golden path. Those numbers go up whether or not the platform is good, because leadership asked people to use it.
There is one measurement that cannot be gamed by mandate. Take a new service, from an empty repository to serving production traffic with monitoring and an on-call route, and time it.
Measure it with a real team, not a demo
The measurement only means something if the person doing it is a normal engineer on a normal team, not a platform engineer who wrote the templates.
Run it as an exercise with someone who has never done it. Watch. Do not help. Every time they stop to ask a question, that question is a gap in the platform, and the list of questions is a better roadmap than any survey you will run.
Golden paths beat golden cages
The failure mode of platform work is a path so opinionated that any team with an unusual requirement has to leave it entirely, and once a team leaves it they never come back.
A good path is easy to join and possible to leave partially. A team should be able to take the pipeline and not the runtime, or the runtime and not the observability defaults, without being told they are off the supported route.
- Every default should be overridable in one obvious place
- Leaving one part of the path should not forfeit the rest
- The escape hatch should be documented, not folklore
The platform is a product with unusually captive users
Captive users do not file feature requests. They build workarounds quietly and complain in their own channels, and the platform team hears about it as a general sense of unpopularity rather than a specific defect.
The remedy is the ordinary product one. Talk to the users, watch them work, and treat a widely copied workaround as a specification for the thing you should have built.
Documentation is part of the interface
A platform whose behaviour can only be learned by reading its source is a platform with one real user, the team that wrote it.
The test is the same as the headline metric: can a new engineer get to production using only what is written down. If the answer requires a conversation, the conversation is the missing documentation.
