One endpoint, on purpose. A lane is a level, and a second endpoint for one concept
is a second vocabulary. Company, partitions and lanes come back together — and
where a tier does not exist in the schema it is reported as absent with the
reason, not omitted.
The tree
recallable_beliefs counts only candidate, active and confirmed claims —
the same set GET /v1/brain/beliefs returns by
default.
lanes: [] means the tier does not exist
An empty lanes array here does not mean “you have no lanes configured”. It
means the concept has no storage. absent.lanes is the field that says so, and it
is always present.The same applies to level: "agent" on the beliefs endpoint: it is not a filter
that returns nothing, it is a value the schema cannot produce. Asking for it is
400 invalid_input rather than an empty list.
partition_model is an open question, reported rather than resolved
The contract models N named partitions per project. The brain schema makes a
partition be a knowledge base. Those are different models and neither has been
ruled on, so this response reports the schema — knowledge-base rows as
partitions — and says in absent.partition_model that it is doing so.
Practical consequence: do not write code that assumes a fixed number of partitions
per project, and do not assume a partition name is stable across a future ruling.
Vocabulary
The DSL levels are company | team | agent. The stored scope values are org and
project. This endpoint maps org → company and everything else → team, and
reports the DSL word on every row. A reader who has to remember which of two
vocabularies a field is in will eventually get it wrong, so only one appears here. 
