Colour, a page you can see the edge of, and a document on its own

Five of the nine things on the list, and one finding that came out of them.

Colour: a marker pen and an ink colour, written as marks here rather than
installed — the schema of a note decides what can be stored in it forever. A
highlight is a <mark> so a reader that drops the colour still shows the word
as marked, and marked text is painted dark whatever the theme. Both travel as
HTML in Markdown, like underline, and both are read back.

The PDF fills the screen: the browser's own full screen for the window frame,
an overlay for the sidebar and the list, Escape and a visible button for the
way out.

The dashboard leads with the recent notes, sorted by time whatever order the
notebooks are in. A tap on the blank part of a page puts the caret at the
nearest position instead of nowhere. A drag from the left edge shows the
sidebar. The page has an edge on a phone, where it had none.

The finding: at the speed a robot types, a long note loses keystrokes between
the editor and the pagination measuring after each one — 87 of 220
paragraphs. Nothing is lost at 50 characters a second, so the page-count
check now pastes its bulk text rather than measuring typing throughput. And
the history suite asked for "the first note in the list" without waiting for
the note to reach the server; it asks by name now.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

e2e/pdf.mjs

@@ -270,6 +270,49 @@ try {
   await p.getByTitle("Zoom out").click();
   await p.waitForTimeout(1500);
 
+  /* ── the document, and nothing else ──
+   * Two things at once: the browser's own full screen, which puts the page past
+   * the window frame, and an overlay over the app, which is what actually
+   * removes the sidebar, the list and the title bar. The second is the half that
+   * can be measured here — and the half a browser can refuse. */
+  const workspaceBox = () =>
+    p.locator("[data-pdf-workspace]").evaluate((el) => {
+      const r = el.getBoundingClientRect();
+      return { x: Math.round(r.x), y: Math.round(r.y), w: Math.round(r.width), h: Math.round(r.height) };
+    });
+  const windowBox = () => p.evaluate(() => ({ w: window.innerWidth, h: window.innerHeight }));
+
+  const normal = await workspaceBox();
+  check(normal.y > 0 || normal.x > 0, "in the ordinary layout the document shares the screen", JSON.stringify(normal));
+
+  await p.locator("[data-pdf-fullscreen]").click();
+  await p.waitForTimeout(1200);
+  const filled = await workspaceBox();
+  const win = await windowBox();
+  check(
+    filled.x === 0 && filled.y === 0 && filled.w >= win.w - 1 && filled.h >= win.h - 1,
+    "full screen leaves the document the whole window",
+    `${JSON.stringify(filled)} vs ${JSON.stringify(win)}`
+  );
+  check(
+    (await p.locator("[data-pdf-position]").count()) === 1,
+    "with its toolbar still there — the page number, the zoom, the way out"
+  );
+
+  await p.keyboard.press("Escape");
+  await p.waitForTimeout(1000);
+  const back = await workspaceBox();
+  check(back.y === normal.y && back.x === normal.x, "and Escape gives the app back", JSON.stringify(back));
+
+  await p.locator("[data-pdf-fullscreen]").click();
+  await p.waitForTimeout(1000);
+  await p.locator("[data-pdf-fullscreen]").click();
+  await p.waitForTimeout(1000);
+  check(
+    (await workspaceBox()).y === normal.y,
+    "and so does pressing the button again — the way out is visible, not only known"
+  );
+
   /* ── Ctrl and the wheel ──
    * The claim is not only "the number changes". It is that the browser is kept
    * out of it — an unprevented Ctrl+wheel zooms the whole application, toolbar