About the Subreddit Atlas

A map of NSFW Reddit communities, positioned by who comments where.

What This Is

This site is a network-analysis project. Each dot is an NSFW subreddit, and two subreddits sit near each other when they share commenters. No text from the communities is used to compute the layout. These communities are image-heavy and text-sparse, so the words they contain carry little signal about how they relate to one another. Who participates where turns out to carry that signal instead: the map recovers regional communities, kink neighborhoods, and language groupings purely from overlapping participation.

How the Map Is Built

For each subreddit I count which authors commented there during the most recent twelve months of data, then binarize that to commented-or-not so that power users cannot dominate. Each author is also weighted by the inverse of how many of these communities they appear in, so that someone active in a hundred of them counts for no more in total than someone active in three. From the resulting author-by-subreddit incidence matrix I compute subreddit co-occurrence, reweight it with positive pointwise mutual information (PPMI) to correct for community size, reduce it to 150 dimensions with a truncated SVD, and project to two dimensions with UMAP. Clusters come from HDBSCAN run on that same two-dimensional projection, so the regions you see are spatially coherent on screen rather than being groups from a hidden higher-dimensional space that happen to scatter when flattened. Each region's label is generated automatically by c-TF-IDF over the subreddit names and public sidebar descriptions it contains. The regions are then grouped into about thirty broad areas by how similar their audiences are. Those are the names shown when the map is zoomed out, and each area has its own color. Because areas are grouped by audience and not by position, an area's regions are usually close together but not always adjacent.

Two thresholds shape what appears. An author is counted only if they commented in at least two of these communities and no more than two hundred — one is not evidence of a connection, and hundreds is a bot rather than a person. A subreddit is mapped only if at least 100 distinct commenters remain after that filter. The two interact, since dropping authors can push a community below the threshold, so they are applied repeatedly until the set stops changing. The map's size filter starts at 250 commenters, because a position that rests on fewer shared commenters is less reliable; lowering it shows the smaller communities.

One caveat worth stating: UMAP preserves local neighborhoods, not global distances, so how far apart two regions sit on screen is not a calibrated measure of how unrelated they are. Adjacency is meaningful; long distances are not.

Data Source and Scope

The underlying data comes from the Arctic Shift monthly archives, distributed through Academic Torrents. This map uses the most recent twelve months available, so it is a snapshot of one year rather than of all Reddit history. The exact window is shown at the bottom of the map, and it moves forward as new months are published; the date beside the title is when the map was last rebuilt. Accounts with bot-like names or bot-like activity patterns are filtered out before anything is counted.

The map covers every public NSFW subreddit that clears the commenter threshold above, so a community's absence usually means it fell below that cutoff, not that it was excluded deliberately. Subscriber counts shown in the detail panel come from a January 2025 snapshot and can be well out of date. Whether a subreddit counts as NSFW comes from Reddit's own over18 flag in a January 2025 metadata snapshot, which has two consequences worth knowing: communities created after that date carry no flag and cannot appear, and any community whose operators never set the flag is invisible here regardless of its content.

Grey dots are communities HDBSCAN left unassigned — they sit between regions rather than inside one, either because their commenters are spread across several neighbourhoods or because they overlap too little with anything else to anchor them. They are on the map and searchable like any other; they simply belong to no labelled region. Their nearest-neighbour lists are often the more useful way in.

Separately, the comment data is a fixed snapshot while Reddit keeps changing, so communities can be banned, go private, or be quarantined after the fact. Every mapped community is rechecked when the site is rebuilt. Any that Reddit has banned or deleted is dropped from the map entirely, along with every reference to it in other communities' nearest-neighbour lists. Ones that still exist but have closed to the public are kept, dimmed, and labelled as private, restricted, or quarantined in their detail panel.

What Is and Is Not Collected

The pipeline reads metadata only: author names, subreddit names, timestamps, and counts. Comment bodies, images, and URLs are never read out of the archives or written to any output. The published map contains no usernames at all. What you can see here is limited to subreddit names, aggregate commenter and subscriber counts, derived 2D coordinates, and each subreddit's public description.

Visitor Statistics

The site counts visits with GoatCounter, which sets no cookies and does not track visitors across sites or visits. It records the page viewed, the referring site, browser and screen size, and a country derived from the IP address, which is not stored. On the map, opening a community's detail panel is counted as well, so the totals show which communities people look at.

Browsing Without the Map

Every mapped community also has a plain page listing its most similar subreddits by shared commenters, reachable from the directory of all mapped subreddits.

Removal Requests

If you moderate a subreddit shown on this map and would prefer it not be listed, email atlasafterdark@proton.me from an account that can plausibly speak for the community, and I will remove it. The map is rebuilt periodically, so removals are applied as an exclusion list rather than a one-time edit.

Supporting the Site

The site is free, has no ads, and sells nothing. If you find it interesting and want to help cover hosting and domain costs, donations in Monero or Bitcoin are welcome. Monero is the more private of the two, and its fees are negligible for small amounts.

Monero (XMR)

42DLobhgu2J9yfjbRGaneGhsYgEQLRbsWVjf6BEv76azgpRXX9atYmD18cBpDjUzTVaQkZ2cpRJZWBbCQpaBDQjkMssjmiL copy address

Bitcoin (BTC)

standard silent payment

bc1qkd0xyt7ta36dg47rtkjqwx5q2vn5na3zachl94 copy address