[jackulau]

What

In the viewer, a page or list shortcut no longer takes a key the focused control uses itself, and a redraw no longer drops keyboard focus onto the page body.

• Enter and Space. Enter on a focused button, filter chip, link or summary presses it, and so does Space on a button. Before, Enter on a filter chip opened the selected session, Enter on a button inside an open case row folded the row, and Space on a replay button started Play.
• Scroll keys. ↑ ↓ PageUp PageDown Home End scroll the step detail when it has focus and more to scroll. j and k still move between steps from there.
• Focus after a redraw. When a redraw removes the focused control, focus goes to the control now in its place (a rebuilt chip, speed button or Explore control), else to its pane or the page. A page change focuses the page. A "show more" button that removes itself hands focus on the same way.

Behaviour change to know about

Clicking inside the step detail focuses it, so ↓ after a click there now scrolls the detail instead of moving to the next step (j still does). To undo only this, drop the SCROLL_KEYS branch in ownKey().

How

• ownKey(e) in index.html is the one rule: the global key handler and each list's Enter/Space handler ask it first.
• swap() now runs through keepFocus(root, redraw), so every redraw that goes through the page's one DOM helper keeps focus. Three direct replaceWith/remove sites use it too.

Tests

test/viewer/keyboard-browser.test.ts: eight tests with real key presses (Input.dispatchKeyEvent, new page.press() in the test driver). The existing tests dispatch KeyboardEvents from the page, which never run a key's default action, so they could not see these bugs. Each new test fails without its part of the fix.

• Viewer and docs tests: 277 pass. Typecheck, npm run build and npm run lint pass.
• Full npm run check on a Mac (run before the list-row fix and the docs rewording): 6257 tests, 6244 pass, 11 skipped, 2 fail. The two failures (test/capture/hook.test.ts "concurrent hooks ... never interleave" and test/cli/rebuild-redactions.test.ts) fail the same way on an untouched origin/main in the same shell; cause not found, and this change does not touch that code.
• coding.sh doctor on a live install: not run (no adapter change).