# EI.7.05 — slide notes beside the deck, in the console

Chromium through the inspection harness
(`apps/oshun/web/e2e-inspect/presentations-notes.spec.ts`), against the admin
dev server on :3020, its own same-origin proxy, the BFF on :4010 and a real
migrated Postgres. `signInAsOperator` from
`apps/oshun/web/e2e-inspect/support/admin-session.ts`, as EI.3.07 uses.

Run:

```
psql "$ADMIN_URL" -c 'CREATE DATABASE eve_notes_e2e'
( cd libs/oshun/persistence && OSHUN_V1_DATABASE_URL=".../eve_notes_e2e" npx prisma migrate deploy )
( cd apps/oshun/bff && OSHUN_PRESENTATION_NOTES_DATABASE_URL=".../eve_notes_e2e" PORT=4010 npx tsx src/server.ts )
( cd apps/oshun/admin && NEXT_PUBLIC_OSHUN_BFF_URL=http://127.0.0.1:4010 npx next dev -p 3020 --webpack )
( cd apps/oshun/web && npx playwright test --config e2e-inspect/playwright.config.ts presentations-notes )
```

`chromium-notes-configured.txt` and `chromium-notes-unconfigured.txt` are the
two runs, unedited: 4 passed / 1 skipped, then 1 passed / 3 skipped.
`chromium-layout-regression.txt` is EI.3.07's own check re-run against this
change, 4/4 with axe clean.

Nothing under these runs is doubled. The proxy is the real one, the BFF is the
real one, the library is this checkout's published centre and the store is
Postgres — because a mocked 200 is exactly what let EI.7.06 ship a delete that
had never worked.

## What the browser found that nothing below it could

**The console never learned which slide was on screen.** Measured on arrival at
`/presentations?guide=eve-meet`:

| | |
|---|---|
| portal posts `oshun-presentation:guide` | 157 ms |
| deck posts `oshun-presentation:slide` | 242 ms |
| (a second announcement round) | 418 / 509 ms |
| React hydrates, and only then is the listener attached | 1350 ms |

Every announcement landed in a window with no listener on it. The deck says
where the reader is ONCE, and `EI.3.02`'s store has no way to ask, so the chip
of EI.3.04 stayed empty for the whole visit, "Ask about this slide" (EI.3.06)
never appeared, and this item's notes list had no subject at all. A `goto`
posted by hand at 9538 ms was heard perfectly — the contract works; nothing was
asking. `?slide=` was deadlocked by the same gap: `PresentationsWorkspace`
waits for a slide before sending the goto that would produce one.

The fix is in `AD/components/PresentationsWorkspace.tsx`: if no slide has been
announced 750 ms after the frame's `load`, the console sends the contract's own
`oshun-presentation:reload`, once per mount, and the deck announces again.
Measured after: announcement 1455 ms, hydration 2150 ms, re-announcement
3046 ms, chip reads `Slide 1 of 11 · Meet Eve` with nothing done by hand.

That is a **repair, not the cure**. A deck that could answer "say where you
are" without reloading itself is the right fix, and it is a change to the
message contract and to the Python builder that generates all 69 decks — filed
as its own item rather than smuggled in here.

## Measured

| claim | how |
|---|---|
| a note written here survives a reload | added on `eve-meet-cover`, fresh document, read back from Postgres through the proxy |
| the guide count is the server's `total` | one note on the slide, two in the guide; the guide read asks `limit=1`, so a panel counting its rows would print "1 note in this guide". It printed `· 1 here · 2 notes in this guide` |
| the stale badge fires on a real slide edit | `takeaway` of `eve-meet-cover` edited in `content/57-eve-meet.json`, the real library; badge reads "Slide changed", `data-note-stale="true"`, the note still readable. Restored byte-for-byte, and the badge goes away again |
| a deleted note stays deleted | deleted through the confirm, gone after a reload, guide count falls 2 → 1 → 0. The rows are tombstoned (`deleted_at`), which every read honours |
| the unconfigured deployment says so | BFF restarted with no `OSHUN_PRESENTATION_NOTES_DATABASE_URL`: 503 `presentation_notes_not_configured`, and the panel says "Notes are not configured on this deployment, so there is nowhere to keep them" — not an empty list, and no composer |
| the new controls are accessible | axe WCAG 2.2 AA with the disclosure OPEN and a note in it: 0 violations. EI.3.07's own axe pass runs with it closed, so none of these controls were inside that signoff |

## What the component spec proves instead

`apps/oshun/admin/src/__tests__/SlideNotesPanel.spec.tsx`, 14 cases, answers the
network with a store that follows the routes' contract — the right boundary for
a component, and the wrong one for the feature. Three negative controls were
run against it: counting the guide from the page length (the count case fails,
printing "1 note in this guide"), reading an absent `stale` as `false` (two
cases fail), and putting a JSON content-type on the DELETE (the delete case
fails, which is EI.7.06's defect exactly).
