# The dockerized rig Two rendezvous relays, three mock chain back ends behind two gateways, and two clients that are told only a relay address. It proves the product path without a real chain: ``` relay-a, relay-b torly with docker/policy.json (two gateway keys registered) node-btc, node-blake, node-alt nostrino-testnode (20, 25 and 30 blocks) gw-a one gateway, one key, two chains (btc, blake) on both relays gw-b a second gateway, its own key, chain alt client-a, client-b on the relay network only; they cannot reach the nodes ``` Run it with `../docker/run.sh` (or `./run.sh` from the repository root). It builds the binaries statically, starts the rig, and then: 1. client A, given only `ws://relay-a:3334`, discovers the two gateways (one listing btc+blake, one listing alt), and syncs **all three chains** through the tunnel to their own node tips; 2. client B does the same from `ws://relay-b:3334`; 3. relay-a is stopped and client B still syncs all three through relay-b (the advertised relay list is what makes that possible); 4. relay-a is started again and client A syncs. The binaries in `bin/` are build output and are ignored; `run.sh` produces them. What it does not cover, and where that is covered instead: the relay's write policy refusals and the market events are unit/relay tests in torly and blake-swap, and the live policy checks were run by hand against swaprelay.smesh.lol.