Composer mic-request + candidate readiness/self-preview (morph without reacquiring media)

[DLSS5ONECLICK]

Problem

Requesting the mic today is a separate, generic control, and camera/mic preview doesn't exist before promotion — a candidate has no way to see themselves or know their setup works until they're already live. Promotion itself re-acquires media from scratch via activateMedia()'s setCameraEnabled/setMicrophoneEnabled calls, which is a second getUserMedia() moment that can flicker/fail independently of the first.

Scope narrowed (2026-08-22)

The composer-integrated request design bullet below is implemented — real-device speaker-entry-friction testing made removing the standalone "Request the mic" control (and its separate justification form) urgent enough to pull forward, alongside #27. This issue now covers only what's left: candidate readiness (local media pre-acquisition), the persistent self-preview, and promotion without reacquiring media.

Scope narrowed further (2026-08-22)

Candidate readiness, the persistent self-preview, and promotion without reacquiring media are all implemented — see the comment below and DECISIONS.md for the full design and reasoning. This issue now covers only what's left: role-specific dominant video and a mic activity indicator.

Design

Composer-integrated request ✅ shipped: the chat composer has a mic-mode toggle (🎤) — submitting in mic-mode posts the request through the existing authoritative requestToSpeak path, replacing the standalone "Request the mic" control entirely. micRequestMode is a controlled prop lifted to EventRoom (not local to the composer), since tapping an empty seat that turns out to have a queue (issue #27) needs to switch this same composer into request mode from outside it.

Readiness without a second gesture ✅ shipped: submitting the mic-mode composer is itself the real user gesture that calls createLocalTracks({ audio: true, video: true }) once (useLiveRoomConnection.prepareLocalMedia), acquiring both tracks under Safari's gesture requirement — one combined permission prompt, not two.

Persistent self-preview, spatially stable ✅ shipped: a new SelfPreview component, always mounted whenever localVideoTrack is held, renders into #20's fixed corner slot, attaching the locally held track via .attach() — the same component/track instance through candidate readiness, promotion, and (per this issue's own explicit non-goal below) still just the local feed once actually speaking.

Promotion without reacquiring ✅ shipped: useLiveRoomConnection's publish path checks for already-prepared tracks first and calls publishTrack() on them directly, skipping a second getUserMedia() call entirely. Falls back to the existing gesture-gated setCameraEnabled/setMicrophoneEnabled path, unchanged, for anyone who never pre-acquired (issue #27's direct join).

Role-specific dominant video (partial, narrow) and mic activity indicator — remaining, unchanged from original design. The former is deliberately deferred rather than built now — the self-preview stays the local feed only; making the other speaker's video dominant once the local user is speaking is a larger visual redesign left to a later pass (possibly #18), not silently expanded into this one.

Explicit non-goal

No candidate snapshot/profile-preview icons for other waiting candidates. No change to #23's automatic-promotion scope — confirmed: resolveClaimDecision (the shared eligibility rule checkPromotionEligibility/claimOpenSeat both use) depends only on queue rank, never on media/readiness state, so this issue's remaining work required no changes there. No new server-side "readiness" field — local track possession is the whole signal (see DECISIONS.md).

Depends on

#20 (video-first room shell) — the self-preview slot must exist. Shipped.

Blocks

Nothing currently — #23's automatic promotion doesn't depend on any signal this issue produces (see the non-goal note above; that dependency assumed a design that changed once #23 was actually implemented).

[DLSS5ONECLICK]

Partially implemented: the composer-integrated request design bullet is merged to main (commit 41accc0), deployed to production — the standalone "Request the mic" control is gone, replaced by a 🎤 toggle on the existing chat composer. Issue body amended to narrow remaining scope to readiness/self-preview/promotion-without-reacquiring, which are not implemented. lint/tsc/build/test all pass (176/176). Not marking Done — this issue isn't complete, and the shipped slice's actual UX needs real-device confirmation per AGENTS.md's new verification-tier rule. Staying in Testing/Review.

[DLSS5ONECLICK]

Remaining scope shipped (2026-08-22, commit e887191): candidate media readiness, persistent self-preview, and promotion without reacquiring media.

• Submitting the mic-request composer now acquires camera/mic itself (one combined createLocalTracks() call, from the submit gesture) instead of waiting for the separate post-promotion "Tap to enable camera & mic" step.
• New SelfPreview component renders the held track into the reserved top-right slot — hidden entirely (not an empty placeholder) with no local media, and the same mounted component/track survives pending → countdown → published-speaker without ever being re-created or reacquired.
• Promotion now calls publishTrack() directly on already-held tracks — no second getUserMedia() call, no second permission prompt. Falls back to the existing gesture-gated path unchanged for Direct join on an uncontested empty seat #27's direct join, which never pre-acquires.
• Deliberately no new server-side readiness field — local track possession is the whole signal. Confirmed resolveClaimDecision (Direct empty-seat join + automatic ranked promotion #23's shared eligibility rule) depends only on queue rank, not media state, so this required no changes to automatic promotion's eligibility logic.
• Neither existing camera/mic-activation control was removed — both remain the correct recovery path for Direct join on an uncontested empty seat #27's direct join and for an acquisition failure (which now also surfaces its own error + retry in the previously-silent pending-request state).

Full reasoning in DECISIONS.md ("Candidate media readiness is a local fact, not a new server field"), session notes in SESSION_LOG.md.

Narrowing this issue's scope further: role-specific dominant video (the other speaker's feed taking over the main stage once you're speaking — deferred, possibly #18) and a mic activity indicator remain open, unchanged from the original design. Not closing this issue — leaving it in Testing/Review pending the user's own real-device confirmation, then will split the two remaining bullets out if/when picked up.

[DLSS5ONECLICK]

Duplication bug found in real-device testing of the previous pass, fixed (2026-08-22, commit `94778e3`): a promoted speaker's own camera was rendering both as their own large tile in the two-seat grid and again in the corner self-preview.

`SpeakerTile` now never shows the big video for the local participant's own occupied seat (`showBigVideo = hasVideo && !isLocal`) — a neutral "You're live — see your preview in the corner" placeholder renders there instead. Presentation-only: track publication and what remote/audience viewers see are untouched, no changes to `useLiveRoomConnection`, no grid/layout resizing. Full reasoning in DECISIONS.md.

Also recorded, not fixed: real-device testing found landscape mode drifting into a dashboard-style layout (small video strip, permanent side-panel chat, full header) rather than staying video-first. Written up as an explicit constraint in ARCHITECTURE.md and on issue #18 (its eventual owner) — no landscape code touched here.

Still not closing this issue — literal asymmetric grid sizing (the other speaker's tile visually enlarged, not just decluttered) and the mic activity indicator remain open, deferred to #18/a future pass. Staying in Testing/Review pending the user's own real-device confirmation of this fix.