# EI.3.07 — the Presentation Center's layout inside the admin console

Chromium, through the inspection harness
(`apps/oshun/web/e2e-inspect/presentations-layout.spec.ts`), against the admin
dev server on :3020 with `signInAsOperator` from
`apps/oshun/web/e2e-inspect/support/admin-session.ts`.

Run:

```
npx next dev -p 3020 --webpack            # in apps/oshun/admin
npx playwright test --config e2e-inspect/playwright.config.ts presentations-layout
```

`chromium-layout.txt` is the run, unedited. 4 passed.

## Three widths, not two

The item names 1440×900 and 390×844. At neither of them can a 960px deck sit
beside a 460px drawer, so a run of only those two would never measure the
side-by-side half of the requirement. 1920×1080 is where it is measurable, and
is included.

`390×844 has no drawer at all.` The copilot's own invocation point declares
`minViewportPx: 1024`, so below that the trigger is disabled by design. "The
drawer overlays" is not something a phone can do, and asserting it there would
have meant either a false pass or a defect reported against a deliberate rule.
The phone is measured for what it does have: the deck at full width, and no
sideways scroll.

## Measured

| | viewport | deck | drawer | side by side |
|---|---|---|---|---|
| wide | 1920×1080 | 1136 | 460 | yes |
| laptop | 1440×900 | 1116 (unchanged) | 460, overlaying | no |
| phone | 390×844 | 358 | none offered | n/a |

The threshold is 1744px = 960 (the deck's floor) + 460 (the drawer) + 324 (the
chrome around the content, measured on this console at 1920). Below it the
drawer goes on overlaying, because a deck squeezed under 960 is worse than a
deck partly covered by a panel the operator can close.

## What the run found, and what was changed because of it

1. **The page did not render at all.** `framedSource` was exported from a
   `'use client'` module and called by the server component: *"Attempted to
   call framedSource() from the server but framedSource is on the client."*
   Every request rendered the shell's error boundary. The component's jsdom
   spec passed throughout and could not have caught it — jsdom does not
   enforce the client/server boundary. Moved to `AD/lib/presentation-route.ts`.
2. **F6 was heard on the way in and not on the way out.** Once focus is on the
   deck the keystroke goes to the framed document, and a listener on the
   console's window never sees it; the second press left focus where it was.
   The handler is attached inside the frame too (same-origin by construction),
   re-attached on each load because the portal swaps decks.
3. **The console had no `prefers-reduced-motion` block.** With chromium
   emulating `reduce`, 44 elements still declared a non-zero transition or
   animation. Added to `AD/app/globals.css` at `0.01ms` rather than `0s`, so
   `transitionend` still fires and nothing waiting on it hangs.
4. **`scrollable-region-focusable` at 1440×900.** `height: 100%` on the
   workspace resolves against the whole content box and knows nothing about
   the workspace title above it, so the deck pushed 54px past the bottom
   (`scrollHeight` 998 vs `clientHeight` 944) and made `#admin-main` a
   scrollable region with nothing tabbable in it. Fixed by making the content
   a flex column so the deck takes the space the title leaves — not by adding
   a tabindex to the scroll container.

Axe reports no violations at WCAG 2.2 AA on either named viewport, with the
framed deck excluded: it is a separate document with its own audit in
`tools/presentations/tests/presentation-center.mjs`, and auditing it from here
would double-report its findings as this page's.

## Negative controls

| control | result |
|---|---|
| raise the 1744px breakpoint out of reach | "side by side on a wide screen" fails |
| delete the `prefers-reduced-motion` block | "nothing animates when the operator asked for no motion" fails |

One more correction to the harness itself, worth recording because it produced
a false reading twice: the drawer must be clicked **after** a wait, not before.
The shell defers its guard to an effect so the first client render matches the
server, which means the trigger is rendered `disabled` and becomes enabled on
the next commit. Clicking immediately measured a blocked button and read like
a blocked route.
