high priority low complexity backend pending backend specialist Tier 2

Acceptance Criteria

Concrete `InMemoryMembershipCache` implements the `MembershipCache` interface defined in task-001
Cache stores `MembershipResolutionResult` keyed by userId in a Dart `Map`
On second call to `resolve(userId)` within the same session, the Supabase query is NOT executed — the cached value is returned directly (verifiable via mock assertions in tests)
Cache is invalidated (entry removed) when `supabase.auth.onAuthStateChange` emits a `signedOut` event
Cache is invalidated when the app calls an explicit `invalidateMembership(String userId)` method (for use by OrgSwitcher after org change)
Riverpod provider for the cache is scoped to the app session — not to a specific widget subtree
Cache does not survive app restart (in-memory only — no SharedPreferences or disk persistence)
Thread/isolate safety: cache access is synchronous and confined to the main Dart isolate
Dart analyzer reports zero errors on new and modified files

Technical Requirements

frameworks
Flutter
Riverpod
Supabase
apis
Supabase Auth State Stream
data models
MembershipResolutionResult
MembershipCache
performance requirements
Cache lookup must be O(1) — HashMap-based
Cache hit must add zero async overhead (synchronous Map lookup wrapped in completed Future)
security requirements
Cache must be cleared on sign-out to prevent stale membership data leaking across user sessions
Cache must not be persisted to disk — membership data is security-sensitive
Cache invalidation on org switch must be synchronous and complete before navigation proceeds

Execution Context

Execution Tier
Tier 2

Tier 2 - 518 tasks

Can start after Tier 1 completes

Implementation Notes

Use the decorator pattern: `CachedMembershipResolver` wraps `SupabaseMembershipResolver` and adds caching behavior without modifying the concrete resolver. This keeps both classes single-responsibility. Subscribe to `supabase.auth.onAuthStateChange` in the Riverpod provider's `build` method using `ref.onDispose` to cancel the subscription when the provider is disposed. Use `ref.listen` or a stream subscription — not a polling timer.

The `invalidateMembership` method should be exposed on the Riverpod notifier so OrgSelectionService (task-005) can call `ref.read(membershipResolverProvider.notifier).invalidate(userId)` after a successful org switch. Keep the cache as a plain `Map` — no need for an LRU or TTL at this stage.

Testing Requirements

Unit tests with `flutter_test`: (1) first `resolve` call hits the mock Supabase resolver — assert mock called once, (2) second `resolve` call for the same userId returns cached result — assert mock called still only once, (3) after `invalidateMembership(userId)` the next `resolve` call hits the resolver again — assert mock called twice total, (4) simulate `signedOut` auth event and verify cache map is empty afterward, (5) simulate `signedIn` event does NOT clear an existing cache entry (sign-in of the same user should not force a re-fetch). All tests are pure unit tests — no real Supabase connection needed.

Component
Multi-Organization Membership Resolver
service medium
Epic Risks (4)
high impact medium prob technical

TenantContextService must invalidate all downstream Riverpod providers when the org context changes (org switch scenario). If any provider caches org-specific data without subscribing to the tenant context, it will serve stale data from the previous org after a switch — which is both a UX failure and a potential GDPR violation.

Mitigation & Contingency

Mitigation: Define a single TenantContextProvider at the root of the Riverpod provider graph that all org-scoped providers depend on via ref.watch(). When TenantContextService.seedContext() runs, it invalidates TenantContextProvider which cascades invalidation to all dependents. Document this pattern in an architectural decision record so all developers follow it.

Contingency: Implement a post-switch integrity check that re-fetches a sample of each major data entity type and confirms the returned org_id matches the newly selected context; surface a reload prompt if any mismatch is detected.

medium impact medium prob security

MultiOrgMembershipResolver must query role assignments across potentially multiple tenant schemas. The anon or authenticated Supabase RLS policy may not permit cross-schema queries, making it impossible to return the full list of orgs a user belongs to in a single call.

Mitigation & Contingency

Mitigation: Design the membership query to use a dedicated Supabase edge function or a shared public schema view that aggregates role assignments across tenant schemas with a service-role key, returning only the org IDs the calling user is permitted to see. This keeps the client read-only.

Contingency: If cross-schema queries cannot be made safely, fall back to a per-org sequential membership check using the list of known org IDs and coalesce results client-side with appropriate timeout handling.

medium impact low prob technical

go_router redirect guards behave differently on web vs. mobile for deep links and browser back-button navigation. If the app is later deployed as a Progressive Web App (PWA) for admin use, the OrgRouteGuard may loop or fail to apply correctly on browser navigation events.

Mitigation & Contingency

Mitigation: Implement the guard as a GoRouter.redirect callback (not a ShellRoute redirect) following go_router best practices for platform-agnostic guards. Write widget tests that simulate navigation with and without auth/org context on both mobile and web target platforms in CI.

Contingency: If web-specific guard behaviour differs unacceptably, introduce a platform check in the guard and apply separate redirect logic branches for web vs. mobile until a unified solution is found.

medium impact medium prob scope

In Phase 2 the OrgSelectionService will need to coordinate the handoff to BankID/Vipps authentication after the org is selected, storing the returned personnummer against the correct tenant's member record. If the service is designed too narrowly for Phase 1 email/password flow, retrofitting Phase 2 will require invasive changes to an already-tested component.

Mitigation & Contingency

Mitigation: Design OrgSelectionService with an AuthHandoffStrategy interface from the start (Phase 1 implementation: email/password, Phase 2: BankID/Vipps). The strategy pattern makes the Phase 2 swap an additive change rather than a rewrite. Stub the interface in Phase 1 with a TODO comment referencing the Phase 2 epic.

Contingency: If Phase 2 requirements diverge significantly from Phase 1 assumptions, create a dedicated Phase2OrgSelectionService subclass that extends the base and overrides the auth handoff step, preserving Phase 1 behaviour unchanged.