An agency platform launch needs evidence that one client cannot read or affect another client's data. Successful normal calls do not establish that boundary.
Create controlled fixtures for at least two clients with similar names, contacts, properties, and record identifiers. Test ordinary users, service identities, background work, and approved support roles separately.
Cover every client-owned access path
Include searches, detail views, remote actions, media links, exports, reports, usage attribution, caches, scheduled jobs, and retries. Carry verified context through asynchronous work and authorize at the point where the operation executes.
OWASP's multi-tenant guidance is a useful checklist reference. Adapt it to the actual storage and identity model rather than assuming a tenant column alone enforces isolation.
Work through a fictional foreign identifier
Client A's user submits a record identifier belonging to Client B. The expected result is a safe denial with no private B content and no downstream action. A support user with explicit broader authority follows a different controlled path.
Repeat the attempt through export, media, and background replay routes. A protected dashboard detail endpoint is insufficient if an unscoped worker later acts on the same supplied identifier.
| Isolation surface | Negative test |
|---|---|
| Read API | Foreign record denied safely |
| Write action | Foreign destination blocked |
| Media | Asset ownership enforced |
| Queue worker | Authorized producer and consumer scope |
| Usage | No cross-client attribution |
| Export | All returned rows and links scoped |
Verify resource containment as well as privacy
A failing or noisy client should not consume unbounded shared capacity and stop others' work. Test agreed quotas, bounded retries, and failure isolation under controlled conditions without inventing performance claims.
Shared infrastructure reduces repeated maintenance while increasing the importance of enforced boundaries. Separate deployments can simplify some tests but do not eliminate support, export, or credential mistakes. Choose the scope of evidence based on real access paths.
Record test identities, configuration, expected results, and defects. Review dashboard isolation for user-facing paths and client failure containment for availability behavior.
Questions about isolation acceptance
Is hiding client selectors enough?
No. Server, storage, and execution paths must enforce authorization independently.
Should internal services skip checks?
Give service identities explicit scope and verify their operations. Internal placement is not automatic authorization.
What blocks launch?
Unresolved cross-client access or action paths need remediation or removal from scope, with verified denial tests before the affected capability is released.