perf(tests): enable isolate: false for unit-core workspace project

[aaight]

Summary

• Enables isolate: false for the unit-core workspace project in vitest.config.ts
• Keeps default isolation (isolate: true) for unit-triggers and unit-backends where complex mock interactions could cause inter-test pollution
• Validated stability across 5 consecutive runs with no intermittent failures

What isolate: false does

Skips per-file module re-evaluation in workers, sharing the module graph across tests in the same worker thread. This directly reduces collect-phase overhead.

Safety analysis

unit-core is safe because:

• Tests use simple mocks (no 18+ vi.mock() chains like backends/triggers)
• All 8 files using vi.useFakeTimers() call vi.useRealTimers() in afterEach/afterAll, preventing timer state leakage
• clearMocks: true in shared config handles vi.fn() cleanup between tests

unit-triggers and unit-backends retain default isolation because:

• unit-triggers: Heavy mock interactions with config-resolver and trigger-check (~37 files)
• unit-backends: Complex mock setups (adapter.test.ts has 18 vi.mock() calls, files that don't restore fake timers)

Performance improvement

Phase	Before	After
collect	42.17s	~40.5s

(5 consecutive runs show stable results: always 156/159 files pass — the 3 failures are pre-existing and unrelated to this change)

Test plan

• unit-core has isolate: false in vitest.config.ts
• unit-triggers and unit-backends retain default isolation
• 5 consecutive runs produce identical results (no flakiness)
• All fake-timer files verified to restore timers in afterEach/afterAll
• Lint passes (biome check)
• Type check passes (tsc --noEmit)

Closes: https://trello.com/c/gna0N4Gc/342-as-a-developer-i-want-test-isolation-disabled-for-pure-unit-test-projects-so-that-per-file-module-re-evaluation-is-skipped

🤖 Generated with Claude Code

🕵️ claude-code · claude-sonnet-4-6 · run details

[nhopeatall]

LGTM - the change is narrowly scoped to unit-core, preserves isolation for the higher-risk unit-triggers and unit-backends projects, and matches the PR description. I verified the touched config and did not find a concrete correctness or architectural issue in the diff itself.

🕵️ codex · gpt-5.4 · run details