TEST-04: Wall-clock sleeps for async synchronization (~90 frontend setTimeout waits, several Go sleeps) #36

Severity: P3. Confidence: high. Informational.

Affected files:

• frontend/src/__tests__/auth.test.ts:126
• frontend/src/__tests__/auth.test.ts:162
• internal/scheduler/scheduler_test.go:1278
• internal/api/ri_utilization_cache_test.go:256-262
• internal/server/integration_test.go:87

Evidence:
43 occurrences of 50ms await new Promise(resolve => setTimeout(resolve, 50)) plus 25 of 10ms and 3 of 100ms across frontend suites, used as fixed waits for async handlers (fake timers are used elsewhere, so the alternative is known). Go: a 20ms sleep to assert a goroutine did NOT fire, 250ms/200ms sleeps against real TTLs, a 100ms "Simulate actual work" sleep.

Impact:
3-5s of dead wall time per frontend run plus CI-load-sensitive flake windows; the 20ms negative-assertion window can miss a slow wrongly-spawned goroutine (false green on the Lambda invariant it guards). The anti-pattern is forbidden by the project's feedback_no_sleep_in_tests rule.

Recommendation:
Frontend: microtask flushes, fake timers, or awaiting the mocked API promise. Go: inject a hook/channel for the negative assertion; fake clock for TTL tests.

Source: docs/reviews/codebase-review-2026-06-10.md (automated multi-dimension code review, adversarially verified for P1/P2)