context.setOffline flips navigator.onLine, but the 'offline' event never fires

Yes. When Playwright sends Network.emulateNetworkConditions { offline: true } over the Chrome DevTools Protocol, Chromium internally flips the navigator.onLine getter to return false for every execution context in the browser context. You can verify it with await page.evaluate(() => navigator.onLine); the value is the boolean false. The call is synchronous from the page's perspective: by the time the setOffline promise resolves, a fresh script eval will read false. The subtle part is that no event is dispatched to announce the change. Code that reads navigator.onLine at a specific moment sees the new value; code that listens for the browser to tell it the value changed waits forever.

Because the entire offline implementation in Playwright's Chromium backend is four lines: crNetworkManager.js:131-137 calls session.send('Network.emulateNetworkConditions', { offline: true, latency: 0, downloadThroughput: -1, uploadThroughput: -1 }). That CDP command affects the network stack, not the DOM event system. Chromium does not piggyback an 'offline' CustomEvent on the state change, and Playwright does not inject one. Any application code that starts its disconnected UI inside a window.addEventListener('offline', () => setState('offline')) handler will keep waiting for a handler call that is never coming, and its banner will never render.

Dispatch the event yourself. The shortest version lives inside page.evaluate right after the setOffline call: await page.evaluate(() => window.dispatchEvent(new Event('offline'))). A more robust version is an addInitScript on the context that wraps context.setOffline so every invocation dispatches the matching event. The page includes a copy-paste Playwright test fixture that does this once and every test in the suite inherits the fix. The cost is three lines; the benefit is that your app's real event handler finally runs under test.

Because Playwright already does that work for you through the CDP call, and overriding the getter is actively harmful. If you defineProperty the getter before your app loads, your override wins and the browser's own offline emulation cannot update it later (setOffline(false) becomes a no-op from JavaScript's point of view). The right separation is: let Playwright manage the value of navigator.onLine, then wrap setOffline so it also dispatches the matching DOM event. The property stays accurate; the event listener finally wakes up.

Yes, symmetrically. Flipping the network back on through setOffline(false) does not trigger window.addEventListener('online') either. If your app reconnects to a WebSocket or refills a stale cache inside an online-event handler, that work never runs under test. The dispatchEvent wrapper must handle both directions: after setOffline(true) fire 'offline'; after setOffline(false) fire 'online'. Either one alone leaves half your offline story unverified.

The Assrt agent explicitly refuses it. The plan-generation prompt at /Users/matthewdi/assrt-mcp/src/mcp/server.ts line 219 reads: 'The agent can: navigate URLs, click buttons/links by text or selector, type into inputs, scroll, press keys, and make assertions. It CANNOT: resize the browser, test network errors, inspect CSS, or run JavaScript.' That exclusion is intentional. Low-level browser state (offline mode, request interception, CDP flags) belongs in a Playwright test you write and own, not in an English scenario handed to an agent. Assrt generates real Playwright, so the offline fixture below drops into the same tests folder and runs next to your Assrt-driven scenarios.

Partly. Service workers hit a separate bug tracked as microsoft/playwright#2311: setOffline did not stop SW-controlled fetches in older versions. WebSocket connections stay alive after setOffline; see microsoft/playwright#4910. navigator.connection.effectiveType does not change with setOffline because CDP's Network.emulateNetworkConditions is distinct from NetworkInformation emulation. Each surface needs its own workaround. The 'offline' event dispatch fix on this page addresses the biggest one, the one most apps actually subscribe to, but an app that also relies on service worker fetch failures or connection.type changes needs additional code.

In a Playwright test fixture. The page shows the full fixture file: tests/fixtures/offline.ts exports a context fixture that patches context.setOffline to also dispatch the event via an addInitScript hook. Every spec that imports from './fixtures/offline' gets a patched context automatically. No per-test boilerplate. You can also drop the script into playwright.config.ts under use.storageState or a custom projects entry if you want the behavior to apply to every project in the config without an import.