- Workspace isolation
- Sessions and API keys resolve a workspace before data routes execute. Membership is checked for selected workspaces; client-supplied identity headers are replaced at the gateway.
- Bounded operations
- Jobs require an idempotency key and explicit maximum units. Batch reservations are atomic. A task with an uncertain data source outcome is held for reconciliation.
- Private storage
- Data source credentials stay in a private server-side service. Result artifacts are stored in private object storage and downloaded through authenticated workspace routes. New-result retention is configurable.
- Team accountability
- Owners manage membership. Admins configure workspaces and keys. Editors run tasks and maintain records; viewers inspect results. Mutations create audit entries.
- Agent boundaries
- External listings, review text and imported records are treated as data. They cannot grant access, change a budget or authorize external communication.
Verification limits
Read the data policy These controls are implemented and covered by selected regression tests. Independent penetration testing, production load validation, legal policies and certified compliance remain separate acceptance gates.