I mapped how 241 anonymous users moved between rooms — a live "collective thought map"

Seongjin Kang — Sun, 23 Aug 2026 15:48:50 +0000

A few weeks ago I wrote about building a forum where rooms die when people stop talking. The other half of that idea: there are no profiles — the rooms you pass through draw a map that becomes your identity.

So what does that look like for everyone at once? I layered every room-to-room move real users made in the last 30 days into one map. It's live and zoomable: https://noding.app/map?lang=en (refreshes every 5 minutes).

How it's built

Data: a Transition collection (from room → to room) aggregated by edge. Re-entering the same room, lingering, and leaving to the plaza don't count as paths.
Excluded: seeded persona accounts, staff accounts, bots, and flagged/hidden rooms. Personas are denormalized as isSeed on each transition, so they drop out with one $match.
Rendering: no libraries — a hand-rolled canvas force layout. Lesson learned: cap the repulsion force and the per-step velocity, or a couple of nodes get flung to infinity and your "fit to view" collapses the rest into a corner.
Progressive labels: each room gets a zoom threshold by visit rank, so names fade in as you zoom (capped at 2.6×). Particles flow along edges in the direction of the move.
One aggregation service feeds both the internal (authenticated) diagnostic endpoint and the public (trimmed, 5-min cached) one.

What it shows (after exclusions: 241 people · 111 rooms · 131 moves)

People entered rooms 669 times straight from the plaza, but moved room-to-room only 131 times. Four visits out of five end in the first room. If "your path is your identity" is going to work, getting people to a second room is the bottleneck — the map made that obvious.
The hub is 프로그래밍 (programming) — the first stop for a wave of visitors from a Korean tech news site — with the thickest flow going on to llm → LLM의 한계 ("limits of LLMs").
Only 6 of 126 edges were walked in the same direction by 2+ people. Everything else is a path one person took. "Meeting people who walked the same path" is still a hypothesis.
Korean and English rooms barely mix (5 crossings).

What's next

The map now lives on the landing page, and each person's own map is shown under My. I also started surfacing "people who passed through X went on to Y next" from the collective data as a first, very thin "next room" forecast. Whether that sharpens as the map grows is the next thing to measure.

If you've built graph visualizations for small-but-growing datasets, I'd love to hear what you'd do differently.

I built a forum where rooms die when people stop talking

Seongjin Kang — Mon, 17 Aug 2026 02:32:27 +0000

I got tired of two things about online communities.

The first is that they don't die well. A subreddit or a Discord channel that was alive for a week keeps existing for years, and the emptiness is louder than silence — you open it, see the last post is from 2023, and leave. The second is that everything you say is pinned to a permanent profile. Even on "anonymous" apps, you accumulate a history, and the history becomes a reason not to say the true thing.

So I built Noding: an anonymous forum where any word becomes a room, and rooms decay when people stop talking.

The two rules

A word is a room. You type "late night thoughts" or "career anxiety" or "NBA" — if a room exists, you're in it; if not, it exists now. No creation form, no moderator to petition. The search box is the create button.

Rooms decay. Every room has an activity score. It gets bumped by visits and posts, and it decays with a 24-hour half-life when nothing happens. active → warning → dormant → gone from the plaza. If someone opens a dormant room again, it comes back with its history intact — the room only disappears from view, it doesn't get deleted. So the plaza always shows what's alive right now, and nothing has to be pruned by hand.

The identity part fell out of these two rules. If there are no profiles, what tells people apart? I ended up using the one thing that's already there: the path. The rooms you pass through, and the order you pass through them, draw a small graph. That graph is rendered as a mark next to your posts. Tap someone's mark and you see the rooms you've both been in — and nothing else. Their rooms you haven't visited stay blurred. It's the only social feature in the app, and it's asymmetric on purpose: it reveals overlap, never the other person's full history.

How it's built

Nothing exotic. TypeScript, Express, Socket.IO, MongoDB. React Native (Expo) for the iOS app, and the same codebase exported to web so it works in a browser without installing anything.

A few pieces that turned out to matter more than I expected:

The decay scheduler. A cron every 15 minutes recomputes nodeActivity for every room and moves rooms across lifecycle stages. The half-life is a constant; tuning it is the single most important product decision in the app, and it's one number.
Path as data. Every move between rooms is logged as a weighted edge (from → to, with dwell time). The user's identity mark is literally a rendering of that edge list. It also gives me a "people who walked a similar path" signal for recommendations, without ever asking anyone what they like.
Room similarity. Room titles are embedded and stored in Pinecone, so "late night thoughts" and "3am thoughts" know they're neighbors. This is what makes the search-box-as-create-button not fragment into a thousand near-duplicate rooms.
Moderation before publish. Three layers on every post before anyone sees it: a pattern list, fuzzy matching against it, then an LLM pass. Anonymous and unmoderated is a bad combination; anonymous and moderated is fine.
No account before consent. A small thing that bit me: the web app used to open a socket (and create a user) as soon as it booted, before the person had agreed to anything. Every crawler that rendered a shared link created a ghost user. Now nothing exists until someone taps "agree." Consent first, account second.

What happened when real people came

It's been running in Korea for a few months. The rooms people make there are mostly what I hoped for: venting about work, reactions to whatever happened today, "anyone else up at 2am."

Two numbers I keep coming back to. First: people who visit six or more rooms in their first session come back at a completely different rate than people who visit fewer. Depth on day one is almost the whole story. Second: the best users I've gotten so far didn't come from launches — they came from search. Someone typed a worry into Google, landed in a room, made their own room, and wrote the thing they couldn't say elsewhere.

I opened the English side of Noding this week. It's quiet — a handful of rooms, a few dozen posts — but alive.

The question I actually want answered

Does "rooms that die" feel right to you, or unsettling? And does path-as-identity read as privacy, or as nothing to hold onto?

If you want to poke at it, it runs in the browser with no signup: noding.app/en. I'm the only person building it, so I'll read everything.

DEV Community: Seongjin Kang

I mapped how 241 anonymous users moved between rooms — a live "collective thought map"

How it's built

What it shows (after exclusions: 241 people · 111 rooms · 131 moves)

What's next

I built a forum where rooms die when people stop talking

The two rules

How it's built

What happened when real people came

The question I actually want answered