[ui] round 5 notes

All nine orders shipped. The tab-click report was a real bug (scrollbar eating
half of every tab hitbox) and is fixed; RENDER's demolition tint is verified
with a screenshot, closing the item owed since round 3.

The honest parts: DATA's commission queue head-of-line blocks the whole game,
RENDER doesn't read the inspect selection yet, and the radio drives a stand-in
until SCREEN publishes its control. One proposal (audio on the bus), no
contract requests.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

fktry/lanes/LANE-UI.md

@@ -471,3 +471,120 @@ breaks for no good reason.
 **NEXT (suggested):** verify RENDER's tint; research rig in reference.ts so a pack economy
 exists to test against; scrolling tab strip if the catalog grows again; era-skinned tech
 panel once eras mean something visually (§8).
+
+### Round 5 (MEGA) — 2026-07-18 — Opus 4.8
+
+**SHIPPED:** All nine orders. New modules: `power.ts` was round 4; this round adds
+`radio.ts` (SPEAKER bridge + echo stand-in), `tuner.ts` (the dock), `screenview.ts`
+(enlarge), and tests `tuner.test.ts`, `wildlife.test.ts`, `screenview.test.ts`.
+**194 UI tests** (was 145); full suite 446/446; `npm run check` clean.
+
+1. **Scrolling tab strip** — one row, never wraps, active tab scrolled into view on
+   `[`/`]`. Found and fixed a real bug doing it: see the hitbox finding below.
+2. **Inspect mode published** — `setSelection({mode:'inspect',entity})` on open, cleared
+   on close (X, Esc, or the unit being demolished under it). Holding a tool supersedes it
+   and shuts the panel, so panel and highlight can't disagree.
+3. **Wildlife surfaced** — infestation toasts in voice; MOSQUITO EXPOSURE + severity in
+   the inspector, read from `snap.wildlife` by `target`; hatch warnings deduped one per
+   pile id.
+4. **Research tests migrated** to SIM's `testkit.grantResearch`/`grantAllResearch` — my
+   round-4 hand-rolled save-blob edit is gone, and with it the coupling to SIM's schema.
+   Tint + highlight verified, see below.
+5. **First-brownout mercy toast** — once per session, on the first brownout event.
+6. **Camera hint** in the idle line; **tab hitbox adjudicated** (below).
+7. **`no seam` jam voice** — "THERE IS NOTHING DOWN THERE. MOVE." SIM emits this reason
+   at index.ts:498, so it's live copy, not speculative.
+8. **THE SCREEN click-to-enlarge** — ~2.5×, dimmed backdrop, Esc/click closes.
+9. **THE TUNER DOCK** — hidden until the fossil is excavated, then slides in with dial
+   (drag + ←/→), AM/FM/SW, volume + mute, and a station readout in voice.
+
+**VERIFIED — live, in the browser** (screenshots taken of each):
+- **Cold-start mercy toast**: placed one extractor with no generator → BROWNOUT, and
+  under it "NOTHING IS GENERATING. THE COMMITTEE SUGGESTS: POWER." That's the DoD's
+  opening requirement.
+- **THE SCREEN enlarge**: the live WebGL sky at 2.5× centred over a dimmed HUD; Esc
+  restored its `style` attribute to `null` and its dock rect to y=10, w=322 — exactly as
+  found.
+- **THE TUNER DOCK**: excavated SIM's real seeded fossil → dock slid in bottom-right with
+  "AUDIO BUS RESTORED. THE BITSTREAM HAS A VOICE.", dial, FM active, VOL/MUTE, hint line.
+- **RENDER's demolition tint — the item owed since round 3 is CLOSED.** Armed remove,
+  hovered a machine, and it renders outlined in demolition red. Screenshot taken.
+- Build bar with padlocks on the POWER page (DECODE ASIC free; SOFTWARE DECODER and ASIC
+  COOLER greyed + 🔒 + REQUIRES:), reference factory healthy at GEN 220 / DRAW 158.
+
+**VERIFIED — headless** (`live.test.ts`, real sim + real DOM, 19 tests): mercy toast fires
+once and *stays* once across a second brownout; inspect selection published on open and
+dropped both on Esc and when the inspected unit is demolished; the tuner stays hidden
+until `relics[0].found` flips and then reveals with the ceremony toast and the toast
+standoff; full research loop (lock → pick → deliver → unlock → place) now driven through
+`grantResearch` and the lab's real per-pack timing.
+
+**DECISIONS:**
+- **THE SCREEN is enlarged in place, not mirrored.** I built the mirror first; it rendered
+  black. THE SCREEN is a WebGL canvas, and `drawImage` from a WebGL context without
+  `preserveDrawingBuffer` returns nothing. Asking SCREEN to enable that flag would cost
+  them a buffer copy every frame for an occasional feature, so I enlarge the real element
+  with inline styles and restore its `style` attribute verbatim on close (tested,
+  including the case where SCREEN later adds inline styles of their own).
+- **The radio bridge is guarded** (`radio.ts`). `src/audio` is still the stub — no `radio`
+  export exists — so the dock resolves a control at runtime and otherwise drives a silent
+  echo stand-in, which is why every dial is testable today. **LANE-SCREEN: publish
+  `window.__fktrySpeaker = { radio }` (or tell me the export name you prefer) and the dock
+  drives real audio with zero change here.** The interface I'm driving is exactly your
+  documented one: `setFrequency / setBand / volume / mute / getState / subscribe`, with
+  `getState()` returning `{freqMHz, band, volume, muted, station}` — `station: string|null`
+  is what flips the readout to "SIGNAL LOCKED". `isLive()` reports which one is attached.
+- Wildlife exposure shows the swarm's **severity** (`WildlifeState.size`), not a
+  sim-confirmed throughput number — SIM owns the real penalty and doesn't expose one.
+  Labelled as exposure rather than a promise about output.
+
+**BLOCKED/BROKEN (read this part):**
+- **The tab-click bug was REAL, and I fixed it.** Reproduced with a genuine mouse click:
+  a click on POWER did nothing. `elementFromPoint` at the button's own centre returned the
+  *strip container*, not the button — the horizontal scrollbar was being painted over the
+  lower half of every tab and swallowing clicks there, which is exactly why keyboard
+  paging still worked. Fixed by giving the scrollbar its own lane (`padding-bottom: 7px`
+  ≥ scrollbar height); the full button height now hit-tests to the button, and a real
+  click activates POWER. **Separately**, `computer` clicks via `ref` deliver *no* events to
+  the page at all (document-level capture recorded nothing); coordinate clicks work fine.
+  So: real bug, now fixed, plus a harness quirk on top of it.
+- **DATA's commission queue head-of-line blocks, and it stalls the whole game.** The v5
+  queue opens with `raw-payload` (20 raw MDAT ORE), but the reference factory refines every
+  scrap of its ore and never ships it — so the head commission can never complete and all
+  19 behind it are unreachable. Probed: 12k ticks ships hf-dust, chroma-slurry,
+  macroblock-bricks, melt and anchor-slab, and `commissionDone` fires **zero** times. My
+  tests now build the obvious player rig (extractor → belt → uplink) to pay it. Worth a
+  design look: either the opening commission should want something the tutorial factory
+  ships, or an unsatisfiable head shouldn't block the tray.
+- **Machines only push onto belts, never into an adjacent machine.** Cost me a while to
+  find; noting it because it's non-obvious and unstated anywhere in CONTRACTS.
+- **RENDER doesn't consume the inspect selection yet** — no `inspect` anywhere in
+  `src/render`. I publish it correctly (verified live: `{mode:'inspect',entity:105}`), so
+  the highlight lights up whenever they read it. **Not owed by me any more; owed to me.**
+- **Stale ghost on keyboard disarm**: main.ts only refreshes the ghost on `pointermove`, so
+  Esc-ing out of remove mode leaves the demolition tint on screen until the mouse moves.
+  Visible in my inspect screenshot. Cosmetic, and it's main.ts's glue, not mine.
+- **SIM landed wildlife AND relics mid-round, again** — my round-start grep found neither.
+  Fifth round running. It's why my wildlife/relic UI could be verified against the real sim
+  at all, so this is a good problem, but the pattern is now completely reliable: I re-probe
+  a *running* sim before writing NOTES, never a grep from round-start.
+- Three of my live tests failed mid-round from SIM/DATA changes, not my code (proved by
+  stashing my diff and re-running). Also hit `DUST_SPILL_GRACE is not defined` from SIM
+  mid-edit, which resolved itself between two consecutive commands.
+- Harness, for the record: the browser pane backgrounds the tab, which pauses rAF **and
+  CSS transitions**. I lost time reading a transitioned `bottom` that never advanced and
+  briefly thought the CSS was broken; `transition:none` proved it computes correctly. Same
+  root cause as the paused sim loop from rounds 3-4.
+
+**CONTRACT REQUEST:** none. v5/v6 gave me everything these orders needed.
+
+**PROPOSAL:** `UIBus.audio?: { radio?: RadioControl }` — the tuner currently reaches THE
+SPEAKER through a global handshake because `ui.init` never sees the audio module, which is
+the same shape of problem as `pickTile` in round 1 and the `__remove` sentinel in round 3.
+One optional field on the bus would make it typed and testable instead of a documented
+convention. Not blocking: the guarded bridge works.
+
+**NEXT (suggested):** wire the real radio the moment SCREEN publishes it (and describe how
+it sounds in NOTES, since they're doing the same); inspect highlight once RENDER reads the
+selection; a station-locked readout screenshot once there are real stations; commission
+head-of-line blocking is a design question for you, not a UI fix.