Each embedded server runs in its own Web Worker that owns a real Indicant instance. Distributors federate by polling every server worker for its snapshot through IndicantClient and answer locate from the snapshots they have pulled. All functionality arrives from Indicant and only the transport is a worker message instead of an HTTP socket, since a static page cannot open ports. Distributors do not have rooms themselves, so every hit is a remote snapshot match.
Cluster configuration
Choose the membership filter, the number of providers (distributors), and the number of servers each provider owns, then rebuild. Rebuilding terminates and respawns the worker servers.
Create a room
The chosen distributor asks one of its own server workers to register a room through the signaling lifecycle hook (indicant.enter), inside that worker. The raw UUID stays in the page; only the double-hashed key enters the filter.
Look up a room
A distributor answers from the snapshots it has pulled. A room created on another provider's server appears only after the next poll, so a fresh lookup can return no hits for up to one poll interval.
Ready.
Live routing accuracy
Locates every created room plus 200 never-created keys against a distributor's pulled snapshots. These filters never miss an inserted key, so missed known rooms should stay 0.
false-positives and phantom hits may rarely occur but that is expected given the probabilistic nature of the filters