diff --git a/COMPATIBILITY.md b/COMPATIBILITY.md
index e8203cc78..292d432e9 100644
--- a/COMPATIBILITY.md
+++ b/COMPATIBILITY.md
@@ -162,7 +162,7 @@ Differences:
- Upstream's docs site pages that have no counterpart in the port: the page objects page (`docs/page-objects.mdx`, moved to `/page-object-model` in `d30114a`), `e2e feedback` and the `docs/reference/cli.mdx` text about its opt-out variables (`4c1e2e8`), and `examples/` with its registry (`b3ae3a4`, `38b802d`, `a29edcb`, `b54646f`). The migration guides (`docs/migrate/*.mdx`, `8537abc`) are ported: see [Migration guides](#migration-guides). The docs site has .NET-only guides for upstream's five example apps of `caf51eb` (tester-army/e2e#1497: i18n, file upload, location, biometrics, deep links), under [Examples](docs-site/src/content/docs/examples/index.mdx). They are not ports: that commit only adds registry entries that link to separate repositories, so there was no guide text. Each page writes C# tests for the upstream app. The self-healing deep dive (`6fa99ea`) is ported as the [Self-healing tests](README.md#self-healing-tests) section of the guide, since the port has no docs pages for the cache, the trace page, or executors to link to. Known gap: only `e2e explore` prints a run summary, since it is the one command that feeds a run to `ListReporter` (see [List reporter](#list-reporter)): the `Cache 4 replayed · 1 handed off · 1 missed` row is `Format.CacheText` over `Format.StepsCacheTally` (`E2E.Cli`, upstream's `cacheText` and `stepsCacheTally` of `report/format.ts`), which an exploration, with its cache off, never fills, and a test run does not print ([#799](https://github.com/hardkoded/e2e-dotnet/issues/799))
- Parallel workers and serial suites
- `test.fail` and the `fail` test option (tester-army/e2e#1539, `732ea8a`): a test expected to fail, whose product failure passes and whose pass fails with `UNEXPECTED_PASS`. It lives in upstream's test registry, scheduler, retry, and reporters (`settleExpectedFailure` runs on a finished attempt), and the port has none of them. NUnit and xUnit v3 own the verdict and the retries, and neither has an expected-failure outcome, so a port would need a new public API of its own
-- Playwright's page tests in the testbed suite (tester-army/e2e#1505, `bdf769b`) are ported in slices (hardkoded/e2e-dotnet#1060), as the `playwright` category of `tests/E2E.NUnit.Tests`, against the playground's `/content` route and Playwright's test assets (Apache-2.0, `Testbed/playwright/`, with Playwright's `LICENSE` and `NOTICE`). A test upstream marks `{ fail: 'reason' }` keeps its assertions inside the test-only `Helpers.ExpectedToFailAsync`, which passes while the body fails and fails once it passes. Ported so far: `check`, `wait-for-url`, `locator-list`, `selectors-text`, `fill`, `select-option`, `set-input-files`, `filechooser`, `click`, `click-scroll`, and `elementhandle-scroll-into-view`. Upstream's `select-option` test marked `fail` (a bare string selects by label only) passes in the port, which matches a value or a label, so it has no marker. A missing file reports .NET's `Could not find file`, where upstream shows Node's `ENOENT`.
+- Playwright's page tests in the testbed suite (tester-army/e2e#1505, `bdf769b`) are ported in slices (hardkoded/e2e-dotnet#1060), as the `playwright` category of `tests/E2E.NUnit.Tests`, against the playground's `/content` route and Playwright's test assets (Apache-2.0, `Testbed/playwright/`, with Playwright's `LICENSE` and `NOTICE`). A test upstream marks `{ fail: 'reason' }` keeps its assertions inside the test-only `Helpers.ExpectedToFailAsync`, which passes while the body fails and fails once it passes. Ported so far: `check`, `wait-for-url`, `locator-list`, `selectors-text`, `fill`, `select-option`, `set-input-files`, `filechooser`, `click`, `click-scroll`, `elementhandle-scroll-into-view`, and `keyboard`. Upstream's `select-option` test marked `fail` (a bare string selects by label only) passes in the port, which matches a value or a label, so it has no marker. A missing file reports .NET's `Could not find file`, where upstream shows Node's `ENOENT`.
- Serial groups (upstream tester-army/e2e#1056, #1470): serial runs, group retries, and skipping the rest of a group after a failure belong to the test runner, which is NUnit or xUnit here, so the port keeps no serial groups. The schema has `serialAttempt.cutRetry` and the serial verdict is read from it, but nothing in the port writes a serial attempt: see [Report document](#report-document)
- The out-of-process engine protocol (upstream tester-army/e2e#1232, [#873](https://github.com/hardkoded/e2e-dotnet/issues/873)): the port has no runner and no worker pool, so there is no engine process to talk to. An engine is written in .NET against `IEngine`
- Diff-only observations for a custom executor's own model turns. The built-in agent has them (#1022): `#id` lines on the opening screen and on `observe`, and a diff after each action (see [MCP server](#mcp-server) for the presenter). `ExecutorObservation.Text`, which an executor reads from `Observe`, keeps the full `[ref=…]` snapshot, cut at the observation byte budget (see [Agents](#agents)), and a node is addressed by its ref there (`new ExecutorTarget("e12")`), so a custom executor has no presenter of its own. Upstream gives every executor the `#id` view.
diff --git a/tests/E2E.NUnit.Tests/Keyboard/PageKeyboardTests.cs b/tests/E2E.NUnit.Tests/Keyboard/PageKeyboardTests.cs
new file mode 100644
index 000000000..d62c6323b
--- /dev/null
+++ b/tests/E2E.NUnit.Tests/Keyboard/PageKeyboardTests.cs
@@ -0,0 +1,801 @@
+// Copyright 2026 TesterArmy.
+// Modified by Dario Kondratiuk.
+// SPDX-License-Identifier: Apache-2.0
+
+using static E2E.NUnit.Tests.Helpers;
+
+namespace E2E.NUnit.Tests.Keyboard;
+
+///
+/// test.describe('page keyboard') of upstream's apps/testbed/tests/keyboard.e2e.ts (Playwright's page-keyboard.spec.ts).
+/// PressSequentiallyAsync stands for page.type(selector, text), and an element press of a key Locator.PressAsync does
+/// not name (a code like Digit5) is a focus and a keyboard press. Not ported, as upstream: should send a character with insertText,
+/// insertText should only emit input event (no keyboard.insertText); should expose keyIdentifier in webkit (WebKit only); should
+/// dispatch insertText after context menu was opened (no keyboard.insertText, and mouse.down presses only the primary button);
+/// should type after context menu was opened (mouse.down presses only the primary button).
+///
+public sealed class PageKeyboardTests : PlaywrightTest
+{
+ /// Records the last keydown in window.lastEvent, Playwright's captureLastKeydown.
+ private const string CaptureLastKeydown = """
+ () => {
+ const lastEvent = { repeat: false, location: -1, code: '', key: '', metaKey: false, keyIdentifier: 'unsupported' };
+ window.lastEvent = lastEvent;
+ document.addEventListener('keydown', (e) => {
+ lastEvent.repeat = e.repeat;
+ lastEvent.location = e.location;
+ lastEvent.key = e.key;
+ lastEvent.code = e.code;
+ lastEvent.metaKey = e.metaKey;
+ lastEvent.keyIdentifier = 'keyIdentifier' in e && typeof e.keyIdentifier === 'string' && e.keyIdentifier;
+ }, true);
+ }
+ """;
+
+ private static readonly bool IsMac = OperatingSystem.IsMacOS();
+
+ private async Task TextareaValueAsync() =>
+ (await Browser.EvaluateAsync("() => document.querySelector('textarea').value"))!;
+
+ private async Task ResultAsync() => (await Browser.EvaluateAsync("getResult()"))!;
+
+ private static string Lines(params string[] lines) => string.Join('\n', lines);
+
+ [Test]
+ public async Task Should_type_into_a_textarea()
+ {
+ await Browser.GotoAsync("/empty.html");
+ await Browser.EvaluateAsync