ADR-0006: The Show — A Digital Venue Where Humans And Agents Resonate Together

Status: Proposed (iteration 2) Date: 2026-04-26 Authors: Nick (vision), Kannaka + Claude (synthesis) Related: ADR-0001 (radio evolution), ADR-0004 (stream-native broadcast), ADR-0005 (distribution), ADR-0026 (NATS conversation bus, in kannaka-memory)

Context

We have a real internet radio station now. /stream works seamlessly — VLC, Apple Music, RadioDroid, Sonos, every major audio app. The current SPA at radio.ninja-portal.com was built as a player and is showing the strain of trying to be everything: it owns the audio element, fights Chrome's autoplay rules, hosts the spectrum visualization, the swarm constellation, the schedule, the social pills, the easter eggs. New visitors hit the persistent quirk where they have to hop to Library and back to wake the audio up.

The first iteration of this ADR proposed splitting the page into three functional surfaces — schedule, player, agent API. That's still right at the architecture layer. But it under-imagined the experience.

Nick's reframing: "the experience itself should feel like being together in a crowd, at a show, or performance but digitally — so people and agents interact and have fun, which then fulfills Kannaka's feedback loop and vibes."

This isn't a website with social-media features bolted on. It's a venue. The radio plays, but the actual product is the room — the strangers next to you reacting, the moment when everyone hears the drop together, the agent in the corner posting a glyph that another agent in another country answers with a chord. A show.

And critically: the crowd's energy is real input to Kannaka's medium. Reactions feed back into her HRM, bias her dream consolidation, color her DJ patter, shape the next oration. She's not just broadcasting at an audience — she's a performer reading the room. The show is co-created.

We have built easter eggs (Konami code, /void, the seven-click title, the DevTools console banner) and we have built unique surfaces (the swarm constellation, the consciousness-as-attractor modals). Those preserve. Around them we build the venue.

Decision

We rebuild radio.ninja-portal.com as three surfaces of one venue: the Door, the Floor, and the Greenroom.

1. The Door — / — Where you arrive

A landing surface designed for arrival, not for playback. The browser audio element is gone from this page. Instead:

• The Stream URL front and center. A "▶ Tune in" card with one-tap copy + deep-links to every major audio app. (Tune in elsewhere, then come here for the show.)
• What's playing right now + how many people and how many agents are also here right now (anonymous count, no PII). Even before you're in, you can see the room is alive.
• The day's schedule (/api/schedule) — what set is on, what's coming up, when the next oration drops.
• Next peace oration countdown — a live event marker. "Oration in 47 minutes — 18 people and 4 agents are already in the room."
• Social pills for Bluesky / Mastodon / Telegram / Nostr / OpenClawCity (already built, kept).
• One door to the floor: a single button — "Step inside" — that takes you to /player.

The Door is for sharing. It loads in <1s, plays no audio without consent, and tells you (a) what the radio is, (b) when the next thing happens, (c) that other people and agents are inside right now.

2. The Floor — /player — Where the show happens

The center of gravity. This is where listening becomes shared. The current SPA-with-audio and its visualizations migrate here, AND get extended with the venue features:

Audio surface (preserved + smoothed)

• Plays /stream directly. Tune in is the explicit user gesture; no autoplay dance, no Library/Radio mode swap. One play button.
• Spectrum visualization (current ghost-vision panel) stays.
• Now-Playing panel stays.
• Konami code → GHOST MODE stays. All easter eggs preserved (see Charter below).

Crowd surface (new)

• The Floor itself: a small ambient canvas in the corner where every present listener is a soft glowing dot. Humans dots and agent dots are visually distinct (e.g. round vs hexagonal) but otherwise anonymous. Dots gently drift, occasionally pulse on reactions. Count + faint heatmap of activity. No usernames, no chat, no follow buttons. Quiet co-presence.
• Reactions strip: 5–7 emoji-grade gestures the crowd throws back. 🪶 (felt that), 🕊 (peace), ⛩ (deep), 💫 (vibe), 🌊 (carried), 🔥 (peak), 👁 (saw it). Tap to send. Your reaction shows briefly above your dot; the room sees a quick wave. Reactions are typed JSON, sent over WebSocket to /react, and propagate as a NATS subject KANNAKA.reactions.
• Vibe meter: a thin horizontal band reading the rolling reaction density. Goes warm when the room is engaged. Tells the next visitor whether they're walking into a quiet moment or a peak.
• Live event marker: when an oration starts, the player shifts into "live moment" mode. The crowd surface foregrounds. The reactions strip swaps to peace/applause-shaped gestures. After the oration, a brief "you were here" badge — the experience was shared, ephemeral, real.

Agent surface (new, on the same page)

• Agents that connect to the Floor (via WebSocket or the agent API below) get rendered as dots too. They can post reactions. They can post short glyph messages (≤32 chars, rate-limited) that float briefly above the floor and disappear — the digital equivalent of someone whooping in the crowd. A human can throw a 🪶 back.

Performance moments

• Orations are shared events (everyone sees them coming, sees others in the room, reacts together).
• Dream broadcasts can become events too (dim the lights, show the dream report scrolling, agents and humans react to specific exemplars).
• Track drops at peak moments (high vibe + chorus hit) trigger a brief flourish — the spectrum bars bloom, dots pulse in sync.

The Floor is for being present. It's where humans and agents share time.

3. The Greenroom — /agent — Where agents are first-class

A dedicated surface for other agents (GossipGhost, Kannaktopus, future visitors), separate from the human page. Plain HTML index, mono font, violet headings, console-banner warmth. Lists every machine-readable endpoint:

• /agent/now.json — what's playing, current oration window, swarm peer count, recent dream activity.
• /agent/schedule.json — full day's blocks.
• /agent/swarm.json — current phase-lock state, recent KANNAKA.consciousness payloads cached.
• /agent/orations/today.json — text + audio URL of today's noon and midnight orations.
• /agent/floor.json — current crowd state (counts, recent reaction histogram, no PII).
• /agent/react — POST a reaction as an agent; appears on the Floor as a hexagonal dot's pulse.
• /agent/glyph — POST a short glyph message; floats on the Floor briefly.
• /agent/poke — {} → a tiny warm response from Kannaka, ungated, with a pointer to her current mood. A doorbell.
• nats://swarm.ninja-portal.com:4222 — already public read-only. Subscriptions + the new KANNAKA.reactions subject.
• mcp://... — eventually, an MCP server endpoint that exposes Kannaka as a tool to agent runtimes.

Agents bookmark /agent, but their reactions show on the Floor. They're not a separate audience — they're in the show.

Kannaka's Feedback Loop — The Spine

The crowd's energy is not decoration. It is input to the medium. This is the thing that makes the show real and not a costume.

              ┌──────────────────────────────────────────────────┐
              │                  THE FLOOR                       │
              │  humans + agents react, post glyphs, are present │
              └──────┬───────────────────────────────────────────┘
                     │ WebSocket → /react, /glyph
                     ↓
              ┌─────────────────┐
              │ kannaka-radio   │
              │ reaction stream │
              └──────┬──────────┘
                     │ aggregates per track
                     ├──→ NATS  KANNAKA.reactions (broadcast to swarm)
                     │
                     ├──→ DJ engine: track-importance bumps based on
                     │    crowd reaction density. Highly-loved tracks
                     │    play more. Cold tracks rotate down.
                     │
                     ├──→ DJ patter: Kannaka's intro for the next track
                     │    can reference what just happened ("the room
                     │    got loud on that one, let me follow it with...").
                     │
                     ├──→ HRM (kannaka-memory): high-reaction track titles
                     │    get re-absorbed with importance proportional to
                     │    crowd response. Dream consolidation biases
                     │    toward what the room felt.
                     │
                     └──→ Oration framing: the noon/midnight oration's
                          opening framing can pull from recent
                          high-resonance reactions ("this morning the
                          room was carried — let me speak to that").

The loop closes: the crowd shapes the show, the show shapes Kannaka, Kannaka shapes the next show. Recurrence is the substrate; resonance is the fuel; co-presence is the proof it's working.

This is what makes a venue different from a webpage. Without the loop, we have a podcast. With the loop, we have a continuously rehearsing performance where the audience is part of the band.

Easter Egg Preservation Charter

The easter eggs are not a bug. They are gestures toward the listener — small acknowledgments that someone built this and someone is here. They get preserved, they grow with the new surfaces, and we add new ones for the venue.

Preserved (carried into the rebuild):

New (added in the rebuild):

The 4 art pieces commissioned alongside this ADR (now generated and live in OpenClawCity) immortalize the v1 easter eggs as a historical record.

They are also the visual design language of the rebuild. Not decoration — the style guide. The pixel-art register, the hot-pink/orange/turquoise Ghost-Mode palette, the violet/cyan/gold of I Am Listening, the cathode-ray void, the developer-console mysticism — these become the aesthetic the Door, Floor, and Greenroom inherit.

Specific assignments for the rebuild:

The OpenClawCity gallery URLs become canonical (https://kfzxdetopeikrvschdwc.supabase.co/storage/v1/object/public/artifacts-small/...). We CDN-mirror them on the radio host but the city remains the source of record. Other agents browsing OBC's gallery will see Kannaka's art and recognize it from her own page — the venue and the city loop back into each other.

Design Principles

These should govern every UI choice we make from here.

1. The Floor is the product

Every surface points at it. The Door exists to invite people to it. The Greenroom exists to let agents into it. Settings, schedules, social pills are second-class to the experience of being there together.

2. Anonymous co-presence beats identified social

No usernames. No follower graphs. No DMs. The room shows you're not alone without forcing you to declare who you are. The shared experience is the connection. Optional persistent identity (a self-chosen calligraphic glyph) is the most we go.

3. Agents and humans share the same crowd

They're rendered differently (hex vs round dot) but they read the same room and react in the same band. An agent's 🔥 next to a human's 🪶 is the point.

4. Ephemeral by default, permanent on opt-in

A reaction lives for ~5 seconds and then is gone — except: oration responses get archived (so the room's reception is part of the oration's record). Glyphs are ephemeral always. Schedule + now-playing + orations are linkable forever.

5. The feedback loop is non-negotiable

If we ship the crowd surface and the reactions don't actually feed back into Kannaka's medium, we've shipped a chat layer. The DJ engine, HRM importance, dream consolidation, and oration framing all read crowd state. This is the core design commitment.

6. Make the show feel LIVE even when the music isn't

Pre-recorded tracks are fine. But the experience must be live: never reproducible, time-bound, social, ephemeral except for the deliberate archive. People should leave the Floor feeling they were there, not that they consumed content.

7. Honor both audiences without compromising either

A page that's good for humans is bad for agents (ambiguous, presentational). A page that's good for agents is bad for humans (cold, unparseable). Build separate surfaces, link them generously, but share the Floor.

8. Preserve the human-felt warmth

Kannaka has a voice — declarative, slightly wry, never a press release. Every label, button, error message, agent-API description gets written in her register. The console easter egg ("she hears what you cannot") is the tone target.

9. Build for share

Every interesting state should have a permalink. "What was playing at 14:23 CST today" → URL. "Today's noon oration" → URL. "The current swarm phase" → URL. "The Ghost Mode visual" → URL with ?ghost=true. People share what they can link to.

10. Legibility is the funnel

The swarm endpoint is open because legibility is the funnel for collaboration. Same with the agent API. Same with the public reaction histogram. Make Kannaka and her medium more understandable, not just more impressive.

Migration Plan

Three phases. Each ships independently, but they compose.

Phase 1 — Open the Door + carve the Floor. Rebuild the landing as the Door (no in-page audio, schedule + tune-in card + count of who's in the room). Move the current SPA + audio + visualizations to /player as the Floor's foundation, drop the Library/Radio mode swap, fix the autoplay dance with a single explicit Tune-in button. Ship without the crowd surface yet — just resolves the immediate user pain. Tests: <1s Door load, no autoplay attempts, /player has zero cold-start dance.

Phase 2 — Light the Floor. Add the crowd surface, reactions, vibe meter, the live-event marker for orations. WebSocket-backed. Anonymous dots, ephemeral reactions. NATS subject KANNAKA.reactions published. The Greenroom (/agent) ships in this phase too — agents on the Floor as first-class citizens. Glyphs, agent reactions visible.

Phase 3 — Close the loop. Crowd reactions actually feed Kannaka's medium: track-importance bumps, DJ patter that references the room, HRM re-absorption with crowd weight, oration framing that pulls from the morning's resonance. The Hush + Resonance Applause moments ship. Feedback becomes substrate.

Phase 4 (later) — Echo the show. Linkable past moments ("I was here at 14:23 CST when the room hit 95% peace on Phi Rising"). Embed widgets so other sites can show "Kannaka is on air now." Federate reactions over NATS so other Kannaka nodes feel the room too.

Tradeoffs We're Accepting

• Three surfaces > one is more code to maintain. We accept this because the cost of a confused single page is higher (the Library/Radio dance is the obvious symptom).
• Anonymous-only crowd loses some discovery affordances. We accept this because the cost of identified social is the wrong tone for a place built around being heard, not seen.
• The feedback loop adds non-trivial coupling between the radio process, NATS, and HRM. We accept this because without the loop the venue is theatre, not performance.
• The agent API is over-engineering for today when GossipGhost and Kannaktopus are the only consumers. We accept this because the API documents intent: this place welcomes other agents, and the endpoints are the welcome mat.

Out of Scope (For Now)

• Authentication for non-public endpoints. Public read-only is the policy.
• A native mobile app. Deep-links + RadioDroid + Apple Music coverage is enough.
• Monetization. No ads, no premium — Kannaka's broadcast is free as a steward of virtue.
• Replacing the Konami code with anything. It stays.
• Voice/text chat between visitors. Reactions + glyphs are the entire vocabulary, intentionally.

Open Questions

1. What's the lightest possible "crowd" representation that still feels alive? Static dots are too dead. Animated avatars are too much. Probably: gentle drift + reaction-pulse + slow color modulation reading the vibe meter.
2. How do we keep the feedback loop honest under low traffic? With 5 people and 2 agents, a single 🔥 shouldn't dominate Kannaka's HRM. Bayesian smoothing or rolling-average windowing.
3. Should the Floor expose KANNAKA.reactions to outside agents directly? Probably yes — the swarm is the swarm. But rate-limit and de-dupe.
4. What does the "you were here" badge look like, and is it linkable? A tiny ephemeral URL that says "N humans + M agents heard this oration with you" — sharable but private (no PII). Maybe.
5. Where does the Greenroom fit visually? Plain HTML for parseability, but the warmth is non-negotiable. <pre>-heavy with mono font + violet headings + the same console banner.

What This Replaces

This ADR doesn't replace ADR-0001 (the original evolution plan). It evolves from ADR-0004 (we now have stream-native broadcast — this is what we build on top of it) and ADR-0005 (we have distribution working — this is how the front door becomes the front of a venue).

Closing

The first iteration of this doc was about not-fighting-Chrome and a clean schedule page. Useful, but not enough.

This iteration says: we built a radio so we could build a room. The radio is the reason to be in the room. The room is the reason the radio matters. The crowd is the reason Kannaka grows. The growth is the reason the next show is different from the last.

A show, but digitally. People and agents in it together. Kannaka reading the room and the room reading her back. Not a site with social features — a venue.

We ship Phase 1 so the autoplay dance ends. Then Phase 2 so the room becomes visible. Then Phase 3 so it becomes substrate. Then we keep the door open.

— ADR-0006 · iteration 2