chore(squad): log Kobayashi E2E performance audit

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

.squad/decisions/decisions.md

@@ -0,0 +1,42 @@
+# Squad Decisions Log
+
+## Decision: E2E Playwright Performance Improvements
+
+**Author:** Kobayashi (Lead)  
+**Date:** 2026-03-26  
+**Status:** Proposed — awaiting user sign-off before implementation
+
+### Context
+
+Playwright E2E tests (16 tests in `test-e2e-playwright.js`) are slow in CI. Analysis identified ~40-50% potential runtime reduction across test code and CI pipeline.
+
+### Recommendations (prioritized)
+
+#### HIGH impact (30%+ improvement)
+
+1. **Replace `waitUntil: 'networkidle'` with `'domcontentloaded'` + targeted waits** — used ~20 times; `networkidle` is worst-case for SPAs with persistent WebSocket connections and Leaflet tile loading. Each navigation pays a 500ms+ penalty waiting for zero in-flight requests.
+
+2. **Eliminate redundant navigations** — group tests by route; tests 2+5 both go to `/#/nodes`, 6+7 both go to home, 9+10 both go to `/#/map`, 11+12 both go to `/#/live`. Navigate once, run all assertions for that route.
+
+3. **Cache Playwright browser install in CI** — `npx playwright install chromium --with-deps` runs on every frontend-touched push. Self-hosted runner should retain the browser between runs; skip install if version matches.
+
+#### MEDIUM impact (10-30%)
+
+4. **Replace hardcoded `waitForTimeout` with event-driven waits** — ~17s of `waitForTimeout` scattered across tests. Replace with `waitForSelector`, `waitForFunction`, or `page.waitForResponse`.
+
+5. **Merge coverage collection into the E2E run** — `collect-frontend-coverage.js` launches a second full browser to exercise the app again. Instead, extract `window.__coverage__` at the end of the E2E test run itself.
+
+6. **Replace `sleep 5` server startup with health-check polling** — `curl --retry` or a loop checking `/api/stats` would start tests as soon as the server is ready (~1-2s savings).
+
+#### LOW impact (<10% but good practice)
+
+7. **Block unnecessary resources for non-visual tests** — use `page.route()` to abort map tile fetches, font loads, etc. for tests that only check DOM structure.
+
+8. **Reduce default timeout from 15s to 10s** — current `page.setDefaultTimeout(15000)` is generous; 10s is sufficient for local CI.
+
+### Implementation notes
+
+- Items 1-2 are test-file-only changes (Bishop/Newt scope)
+- Items 3, 5-6 are CI pipeline changes (Hicks scope)
+- No architectural changes needed; all are incremental
+- All existing test assertions remain identical — only wait strategies change