# UI work — remaining, measured

What is left after the first pass on the two operator surfaces. Each item was
observed on a real render, not inferred. Kept as a list so a gap is recorded
rather than remembered.

## Device console (`bfos/ui_gui.py`)

Reviewed against a screenshot of the running window (`spectacle`, 1440x900),
with the token values read back out of `bfos/ui_theme.py`.

**Landed** — grouping into headed regions; the privilege state moved out of a
monospace paragraph into the header and a banner; the duplicate `Çık` removed;
the output well styled as read-only; status colours from the shared tokens. The
theme file carries the same values as `web/static/tokens.css` (`#1d4ed8`,
`#15803d`, `#b45309`) — verified by comparison, not by assertion.

**Still true at the time of writing:**

1. **The widgets still read as default ttk.** Inputs, comboboxes, checkbuttons
   and the scrollbar keep the renderer's native chrome. `ui_theme.py` selects
   `clam` because it honours `background`/`bordercolor` on the elements it
   restyles, but a themed *element* is not a redesigned control. Closing this
   needs `Style.element_create` work per control, or hand-drawn widgets, and it
   is the single largest remaining gap between this console and the web panel.
2. **The product name appears twice.** `root.title("Blueforce Field OS — Yönetim")`
   and the in-window header band carry the same words. One of them should go;
   the window title is what the taskbar and window switcher show, so the header
   band is the better candidate.
3. **`Cihaz durumu: YOK` is ambiguous** when no target is selected: it reads as
   a state a device is in rather than "nothing selected yet".
4. **`HAZIR` in the action bar looks like a disabled button** — it is a status
   label, but it is styled in the same register as the controls beside it.
5. **The operation list has no unambiguous selected/disabled distinction.** The
   first row carries a focus ring while being unavailable; the ring reads as
   "actionable" and the reason is only in the row's trailing text.

## Central panel (`web/`)

Reviewed against served HTML and captured pages at 1440x900.

**Fixed in this pass** — the accent and the `KURULDU` state were the same blue
(`--state-provisioned: var(--info)`, and `--info` equals `--accent`), so a
selected row and a state badge were indistinguishable; both now resolve to
different colours, asserted by a test that resolves aliases before comparing.
The wave cell printed the bare code and then a label that already contains it
("P2 P2 — PILOT-2"); it now names each wave once.

**Still true:**

1. **No scroll affordance on the device table.** The page reports 13/13 while
   only part of the list is in view, and nothing signals that more rows exist.
2. **The read-only privilege is repeated** in the header note and a second line
   at the far right; one is enough.
3. **The `HAZIR` count appears in the loud figure and again in the breakdown.**
   Defensible (summary then detail), but worth a decision rather than an
   accident.
4. **Timestamps are ISO-8601** (`2026-09-16T06:20:00+03:00`). Precise and
   sortable, but slow to scan for a human at a glance.
5. **A `HATA` device offers no next action** — the panel says a person is needed
   and stops. For a fleet panel this is the workflow gap to close next.