The Deck vNext — scale, spatial grammar, VCM, and the Jarvis line

Status: PROPOSAL — nothing built, nothing changed. Your decision package.

Live deck today: portfolio front door · Mockups: 1 Stage · 2 Compass · 3 Grid · 4 VCM · 5 Jarvis

---

The one idea underneath all eleven asks

Your asks are symptoms of a single missing thing: the Deck is a viewer, not a place. Cards render at fixed size in empty space; swipes go to destinations you can't remember because they're arbitrary; launching is a hidden double-click; the admin tools live in menus instead of your hands. vNext makes the Deck a place with physics: everything scales to the screen it's on, every direction has a permanent meaning, every card knows its medium, and your hands (or finger) use one grammar everywhere — touch, grab, move.

Your asks → what vNext does

• Cards don't scale → fluid stage: the focused card sizes to ~92% of any viewport, phone to tablet, portrait to landscape. One layout engine, no fixed pixels. → mock 1
• The leftover artifact (red circle) → it's the orphaned launch-shelf remnant above the focused card. Deleted outright in vNext; launch affordance moves ONTO the card. → mock 1
• Launching is illegible → a persistent LAUNCH pill on every card with three visible states: ▸ LAUNCH → ◌ OPENING… → ● LIVE. Double-click still works; now it's discoverable, tourable, and identical on all devices. → mock 1
• Can't remember left/right → a fixed spatial map + compass HUD: touch the screen and the edges glow with their destination labels. Left = Signal, right = Command, down = Jarvis, up = lights-out. Same forever. → mock 2
• Swipe → = Command, admin-only → yes, and stricter than "not enabled": for non-admins the right edge doesn't light and the gesture is inert — the feature is invisible, not locked. Admin check = the existing admin cookie.
• Swipe ← = Signal → personal + public as layered cards on phone (tap to flip focus), side-by-side in landscape/tablet. → mock 2
• Grid view (experimental) → Deck Grid: pinch-out from the Stage → many cards fill the screen, featured projects span 2 columns, density slider. Pinch-in returns. Stage remains the public default. → mock 3
• How do we scroll (grid, and sites)? → the dot cursor becomes the finger: touch = hover, grab = drag-pan with momentum, flick = inertial scroll, double-pinch = zoom card to Stage. The exact same verb later scrolls inside launched sites: grab the page, move it. One gesture, three payoffs. → mock 3
• Lights-out needs work → true poster mode: all chrome fades, the card grows to 100% full-bleed, voice pod dims to 25% but stays touchable. Same renderer powers the screensaver feed. → mock 1
• Talk/listen/mute more prominent → the voice pod becomes the single largest control on screen (64px+, center-bottom, state ring: gold pulse = talking, green = listening, red slash = muted). → mock 1
• Font/UI scale control → one A− / A+ control (it had been dropped in a redesign — it returns) driving a single --ui token that scales fonts AND controls: chips, stack pager, star rows, filter pills. Persists per device. → mock 1
• VCM / minority report → ships first as the admin sort room: grab cards in space, drop into FEATURE / KEEP / PARK / RETIRE buckets; writes the existing curation overlay (applies next deploy — no new backend). Public spatial browsing only after the physics prove reliable. → mock 4
• Jarvis line (Gia one-to-one) → swipe ↓ (admin): direct chat with Gia, Mia selectable, mirrored ↔ Telegram by construction (it's a second client on the same conversation, not a sync job). Noise solved with per-topic VERBOSE / SUMMARY / IGNORE filters — generalizing the already-shipped Telegram noise contract. → mock 5
• Talking avatar → staged honestly: A) presence orb (amplitude-driven, cloned Chatterbox voice — ships with the line, 100% reliable), B) viseme portrait (stylized mouth/blink), C) rendered talking head (LivePortrait-class, pre-rendered briefs first; real-time only after a local demo proves it). No promises ahead of reliability. → mock 5
• Media types → cards get a medium: APP ▸ LAUNCH, TALK/VIDEO ▶ PLAY (your re:Invent session, YouTube embeds, chapters), ESSAY ▸ READ (in-deck reader), MUSIC/AUDIO ▶ LISTEN, SLIDES ▸ PRESENT, GALLERY ▸ VIEW (powered by LibesoMedia share bundles — revocable, already built). Registry stays generated; medium is one new field. → mock 3
• Culling/featuring, cleaner controls, consistency → the sort room IS the culling tool; the compass IS the consistency; the pill/pod/A± ARE the cleaner controls. Curation lives in your hands, not in config.

The gesture grammar (complete)

Tours & integration

• Tour engine rule: every view registers its 20-second guided moment (Stage → compass → grid → pill). The tour uses the same labels the compass shows — words match, memory sticks.
• Command / Signal: swipe destinations are the live sites, receiving a ?from=deck param so they can render a return handle (and Command updates per the standing single-pane rule).
• LibesoMedia: GALLERY cards consume share tokens from the share plane — culled sets become portfolio galleries in one step.
• Analytics: every new gesture emits to the existing beacon (grid entered, VCM sorts, compass touches) — so we measure whether the grammar teaches itself.
• Screensaver feed: lights-out renderer is reused as its display mode.

Prioritized build plan

Deployment sequence per phase: build behind a ?vnext=1 flag on the live deck → self-test on phone + tablet viewports → analytics beacon check → tour updated → flag flipped default → registry/CHANGELOG entry. P0 ships alone first (it's invisible-risk, instant quality). Each later phase flips only when its beacon data shows the previous one is being used, not just shipped.

Self-critique — am I being bold enough?

Where the plan is deliberately conservative: Stage stays the default view (Grid is opt-in); VCM debuts admin-only; avatar Stage C is gated on a demo. I stand by all three — the deck IS the brand, and a janky spatial UI or a glitchy talking head on the front door costs more credibility than it earns. Public boldness must be earned in private.

Where I pushed past your ask: the gesture grammar as ONE system across Grid/VCM/site-scroll (you asked for views; the unified verb set is the real product); Jarvis as a second client on the Telegram conversation (kills the mirroring problem structurally); media cards as first-class registry citizens rather than skinned links; galleries plugging into LibesoMedia's existing revocable share plane.

The genuinely bold bet I recommend: commit to the grammar, not just the screens. If touch/grab/move works everywhere by P3, the deck stops being "a portfolio site with gestures" and becomes the thing people remember — and VCM's public debut (P5) is a category of one.

Self-critique — Fable adversarial pass

(Run in-session on Fable 5 — the metered API key has a zero credit balance, flagged to Byron below.)

Brand coherence — the mocks promise more than the plan guarantees. The proposal's aesthetic is coherent, but the brand risk isn't in the visuals — it's in the compass. The moment "left = Signal" ships, that promise is permanent; re-mapping directions later would burn exactly the muscle memory this exists to build. Byron should treat Decision 1 as the only irreversible one on the list and sleep on it, not approve it in a batch.

Gesture reality — three frustrations the mocks hide. (1) Edge-swipes collide with browser back/forward gestures on iOS Safari; the plan needs an explicit answer (inset swipe-zones or press-then-swipe) or Signal/Command will trigger accidentally and feel haunted. (2) The dot cursor is beautiful on a wall demo and clammy on a couch — hand-tracking in poor light will miss grabs, and a missed grab in the sort room mis-files a card. The finger fallback isn't a fallback, it's the primary input 80% of the time; design it first. (3) LAUNCH-pill states depend on knowing when a launched site is "live" — cross-origin, you often can't. Define the honest degraded state now (OPENING… times out to LIVE?) or the pill teaches distrust.

Sequencing — where this dies half-built. P4 (Jarvis) is the highest-want, highest-risk item and it's sequenced last-but-one; if P2/P3 slip, the most differentiating surface never ships. The mirror-by-construction claim is real only if the Telegram bot's conversation model actually supports a second client cleanly — prove that with a 1-day spike BEFORE committing the phase order, because if it needs a relay, P4 doubles.

Boldness verdict — and the tension, resolved. Stage-default is NOT the timid choice — it's correct; the front door must load fast and read instantly for a first-time recruiter on a phone. The timid choice is hiding Grid behind pinch-out only: pinch is undiscoverable. Ship Grid with a visible view toggle AND the pinch. What would make this category-of-one is not Grid or even VCM — it's the pledge that one grammar (touch/grab/move) spans portfolio, admin, and conversation. The proposal says it; the plan should measure it (beacon: % of sessions using grab-scroll by P3) and kill features that break the grammar.

Three highest-leverage changes: (1) spike the Telegram second-client question this week — it de-risks the crown jewel for a day's work; (2) design the finger path as primary everywhere, hands as the delighter; (3) add the iOS edge-gesture mitigation to P1's definition of done.

One bold idea missing: the Deck should have a pulse. Cards subtly reorder by recency of real activity (last deploy, last article, last talk) so returning visitors see a living body of work, not a museum. That's one registry field and it makes every future visit different — cheap, and nobody's portfolio does it.

Decisions I need from you

1. Approve the spatial map (left=Signal, right=Command🔒, down=Jarvis🔒, up=lights-out) — or re-assign directions now, before muscle memory forms.
2. P0+P1 go? (artifact, scale, lights-out, pod, A±, compass, pill) — recommended immediate.
3. Grid density default on tablet: comfortable (~4 col) or dense (~6 col)?
4. VCM sort room in P3 — go/park?
5. Jarvis line in P4 — go/park? If go: confirm default filter levels (costs+deploys VERBOSE, rest SUMMARY, cron IGNORE).
6. Avatar ceiling for now — Stage A only, or greenlight B alongside?

Links