Point
Maintenance cost can defeat Atlas
Identity search, Map routing, context writing, lifecycle review, validation, and policy evaluation must not cost more than the gains they produce.
- Posture
- asserted
- Lifecycle
- active
The semantic layer creates recurring work: authors must search identities, choose Map and Area questions, write real context rather than copies, update canonical state at the anchor, preserve evidence, evaluate Checks, and review stale or superseded records. Tooling can reduce friction but can also hide poor inferences.
Evaluation must count that effort alongside reader and agent gains. A model that improves one cross-context task but makes common authoring or document retrieval substantially harder may still fail. The project should simplify structure before assuming that more automation is the answer.
Costs and failure modesAuthoring and governance overhead belongs in the measured cost of Atlas, not outside its useful-test score.
Learning agendaCost measurements should determine which structures to simplify or remove when their gains do not compensate.