Testing

Lokus has two layers of tests, and they are not equal: the QA harness is the gate. Unit tests check pure logic; the harness drives the real frontend against real temp vaults on disk and reports what actually breaks.

The QA harness (qa/)

npm run qa              # everything: scripted journeys + AI user + monkey pass
npm run qa:journeys     # deterministic journeys only
npm run qa:ai           # AI user + monkey only
node qa/run.js --filter 10   # one journey
node qa/run.js --headed      # watch it drive the app

The harness starts the real Vite frontend with a disk-backed Tauri mock: every file operation hits a real temp folder, so persistence, autosave, and data-loss behavior are tested for real, not mocked away.

Each journey is a scripted user story with screenshots and expected-vs-actual assertions:

Journey	What it proves
00	The client bundle imports no Node builtins (a past crash)
01	Typed text survives a crash/quit without manual save
02–05	Heading input rules, save-to-disk, slash menu, wiki links
06	Split panes keep unsaved content and undo history
07–09	Graph hotkey, window lifecycle, tab isolation
10	Plugin system v3 end-to-end: discovery → Ask Screen → activation → slash/palette/panel slots → worker RPC round-trip → marketplace handling

Results land in qa/reports/run-<timestamp>/ as a self-contained report.html with screenshots per step. Known-broken behavior is listed in qa/baseline.json; any new failure exits non-zero, so the harness is CI-wireable.

Writing a journey

A journey is { id, title, expected, run(d, t) } where d is the driver (createVault, newNote, type, openSlashMenu, click, screenshot, …) and t.expect(cond, name, expected, actual, shot) records a finding. Prefer journeys over unit tests for anything a user can touch.

Unit tests (Vitest)

npx vitest run

~2,000 tests over pure logic: manifest validation, the plugin capability gate, RPC protocol, shortcut registry, markdown pipeline pieces, stores. They run in jsdom with the same Tauri mock philosophy.

What we don’t test with unit tests, on purpose

Tests that only exercise mocks against themselves give false confidence — the suite was once fully green while the app had eight real bugs. If a behavior matters, it gets a journey.

Rust side

cargo check gates the backend in CI; release builds compile on all three platforms per tag.