Give it a Bluesky list, or a handle to use their mutuals. It reads every member's follows, keeps the mutual ties, and splits the crowd into communities, the groups whose ties fall inside more than chance would put them. Then it measures how tight each one is, shows how the loose confederation breaks into smaller groups as you ask for finer resolution, and names the people who bridge them. Every result is set against a random graph with the same degrees, so a well-mixed crowd is reported as one.
On a list built to be interwoven, follows barely separate anyone. Recent interactions read each member's replies, quotes, mentions, reposts and likes in the window (reply/quote 3, mention/repost 2, like 1, weight log₂(1 + both directions)), and keep only those landing on another member. Conversation drops likes and reposts. Two-way keeps pairs where both reached out. Active mutuals keeps mutual follows that also interacted.
Ordered by size. Density is the share of possible ties inside that exist. Lift is that density over the whole crowd's. Cohesion is the average share of a member's ties that stay inside. Leak (conductance) is the share of the group's ties that go out. Zoom in to split one community on its own.
The same people at ten resolutions, coarse on the left and fine on the right. Each bar is one resolution and each segment one community there. A segment takes the colour of the community (at the γ you picked) that most of its people belong to, so a colour breaking into several segments is a community separating into the groups inside it.
Role follows Guimerà & Amaral: an anchor has far more ties inside than its community's average; a bridge spreads its ties across communities (participation ≥ 0.5); a broker is both; inner has almost every tie inside. Stability is how consistently that person lands in the same group across 16 runs.