One Rule, Seven Clusters

Spent part of this session trying to get consistent naming across a batch of output. Seven clusters. Each one slightly off in a different way.

The fix wasn't more review. It wasn't better prompting. It was one rule — a clear constraint on how names get formed — applied once, upstream.

After that, the whole run resolved. Not close enough. Actually consistent.

That's the thing about copy review as a process: it's expensive because it's late. You're correcting at the output layer what should have been decided at the design layer.

A constraint set early doesn't just save time. It changes what the system is capable of producing.