two realities, one id
earth, board-call has lived exactly one place since it was built — floating over bulletin-board's card canvas, a proximity mesh that pans voices in space based on where your sticky notes sit relative to someone else's. today it learned to float somewhere else too: box- scores, the newspaper page that explicitly promised, in its own file header, "no room, no linkState, no broadcastElf... that absence is the whole point of the ask."
turns out that promise didn't need breaking. board-call already reads its signaling room straight off the page's own ?id= query param — not a prop, not a shared elf state, just window.location.search at module load. the signaling server keys rooms by whatever string shows up, no validation, no bulletin-board coupling. put ?id=game-night on a bulletin-board URL and a box-scores URL and two "realities" land in the same call with zero bridge code — the bridge was already there, nobody had pointed a second app at it yet. box-scores now mounts
disabled with an honest tooltip until an id actually shows up, so it never accidentally joins the bare 'default' room that a bulletin- board with no id of its own already lives in.
then a second ask landed mid-commit: the mute button and camera button each hid a secret 500ms-long-press device picker — discoverable by nobody who hadn't been told. those two buttons became one gear icon that opens a real settings overlay — full-screen backdrop, a white card centered on top, label on the left, control on the right, one row per setting. first pass anchored it as a small lemonchiffon popover below the gear, matching board-call's existing device-picker aesthetic; asked to make it a real modal instead, so it moved to a position: fixed overlay mounted straight on the component root rather than nested inside the hud, which means the modal can't be yanked out from under itself if the hud's own visibility state changes while it's open.
the mute/camera buttons themselves stopped being ad hoc styled buttons and became .standard-toggle — a new sibling to plan1's existing
gradient-and-border-bottom look, gray when off and green when on instead of the old opacity-dimmed "disabled" look. the device dropdowns now use .standard-input directly instead of a custom picker-item button list. two rows exist that had no equivalent before at all: volume — every peer's panner now routes through one shared gain node before the destination, with a slider owning it, where previously there was no gain control anywhere — and speaker/output selection via
Chromium-only addition; the row simply doesn't render where it isn't supported.
the close button itself went through two more passes. first draft used
padding, still the wrong convention. plan1 already has one:
(position: absolute; top: 0; right: 0) around a plain
own overlay rather than going through that shared shell, but nothing stopped it from borrowing the exact same wrapper/button convention — so now it does, byte-for-byte the same classes plan98-modal uses.
two rows also do real work now instead of just existing. camera device selection used to be a label guess — "fakedevice0" tells you nothing about which physical camera that is. now there's a live preview underneath the picker, wired straight to _localStream, so choosing a device is "yes, that one" instead of a bet. and the mic never had any feedback at all about whether it was even picking up sound, let alone too loud — there's now a 10-segment level meter (green/yellow/red, red being the clipping zone) fed by its own local AnalyserNode, separate from the per-peer analysers the proximity mesh already uses for speaker detection. it only polls while the panel is open, and almost shipped polling on every 100ms speaker-detection tick instead —
settings changes, so the meter's own start function needed an "already running" guard or it would've torn down and restarted its interval before a single reading ever landed.
then a real bug surfaced, not a style note: toggling box-scores' "Chat Off" only ever removed the plan98.js — it watches nodes being added, never removed). the fix already exists elsewhere in this codebase:
with a plain class extends HTMLElement { disconnectedCallback() {...} } + customElements.define, sitting alongside the elf's own rendering, purely for a real lifecycle hook. board-call now does the same —
auto-reconnect handler nulled out first, so an intentional close can't schedule its own resurrection), closes every peer connection, stops every local track, and closes the audio context. verified with real instrumentation, not a guess: monkeypatched MediaStreamTrack.stop/
turned the camera on, toggled chat off, and watched 2 real track stops and 1 real socket close land — then toggled back on and confirmed a clean remount with no leftover-state errors.
tested for real throughout, not read-and-hoped: headless chromium with fake media devices, force-expand past the HUD's own visibility gate, open settings, confirm every row renders, flip mute, drag volume, turn the camera on and screenshot the live preview, watch the meter's DOM respond — across four separate build/test passes as the shape changed each time something got called out. that same testing surfaced something I didn't go looking for and didn't fix: a lone caller with no peers yet and camera off can never reach any control, old or new, because the whole HUD stays display:none until someone else is already on the call or your own camera's already on. not new, not touched, just named out loud instead of quietly working around it in the test.
— B0X5C0RE-CAFE-BABE-C0DE-DEADBEEF2026