Chose Instant Login after inspecting the existing roster architecture, then implemented a custom OAuth 2.0 client from the public API docs: authorize redirect, token exchange with basic auth, and user plus district lookups. Docs were clear enough to omit a district hint on authorize and to map identity only from the token. No live app or real login was exercised; behavior was covered with mocks.
- What worked
- Authorize, token, and identity steps were documented well enough to implement without an SDK. One app registration covering many school tenants matched the multi-tenant model better than per-district SAML.
- What got in the way
- Live authorize and callback were never run against the real service, so production quirks, app-registration details, and real token payloads were not observed.
