The Docker-in-Docker daemons for Rust runners 1 and 2 retained unused Gitea Actions task bridges. Docker eventually allocated 192.168.0.0/20, which contains the Gitea registry at 192.168.1.4. The pods then routed registry traffic into the empty bridge and build_dev_images failed with connect: no route to host.
After confirming no active tasks on the affected runners, removed only GITEA-ACTIONS-TASK-* networks with zero endpoints: 17 from runner 1 and 16 from runner 2. The registry route and TCP 443 connectivity were restored on both. PR #6688 (1720b4d5) passed branch validation and merged into dev (912997575abf). The repository-managed gitea-act-runner-rust-pool.yaml sets Docker's default address pool to base=10.200.0.0/16,size=24 on all three Rust runners. The live manifest diff changed only that daemon argument.
The three runners were updated individually after checking Gitea for active tasks. Runners 2 and 3 required immediate removal of their terminating old pods after their tasks had stopped because act_runner held its 20-minute shutdown grace; no active task was interrupted. Each new daemon allocated a disposable test network in 10.200.1.0/24, retained the LAN route to 192.168.1.4, and connected to registry TCP 443. The test networks were removed.
Gitea Actions dev run 8802 (index 3721) completed successfully: build_dev_images and rollout_dev passed. autocoder-dev-engine and codecartography-dev-frontend both run image dev-912997575abf with one available replica. The restricted agent launcher authenticated against https://dev.codecartography.dev and listed TheWell SuperProject's 56 ReviewRequired findings.
Revert PR #6688 in Git, verify runner idle state, and apply the previous manifest one runner at a time. The removed zero-endpoint task networks are disposable and require no restoration. The CodeCartography access change has a separate release lifecycle; this record does not imply production deployment.