Interview → prove → apply.
An agent that interviews you, pushes back like a senior modeler, then designs a Lattice ontology and proves it against a live throwaway sandbox — every competency question backed by a query the engine actually ran, every guardrail by a write the engine actually rejected.
Stands up a workspace directly from a prebuilt ontology — no design loop. Uses the slug above.
pick a template and a slug, then create a pre-seeded workspace
Working…
Ontology
Class model
the proposed classes appear here
Relationships
relationship types appear here
Linter
static DDL analysis · one semantic passmodeling antipatterns in the proposed DDL appear here
Question verification
deterministic gate → intent judgeeach question gets a verdict — answerable, or the exact gap
Competency questions
real queries, real verdictseach question is run against the sandbox and shown here
Guardrails
the engine’s own rejectionsillegal writes the engine must reject appear here
Push-back
senior review before modelingApprove → go live
seed the reserved workspace with the proven ontology
Completing the session replays the exact validated ontology into the real workspace at the
slug you reserved, and tears the sandbox down. It is atomic — on failure nothing goes live and
the session stays open so you can retry. Discard releases the sandbox and frees the slug.
approve to provision the reserved workspace, pre-seeded with this ontology
Apply to existing
diff → confirm → versionpreview the diff against your persistent Lattice, then confirm to apply