modules/match-cluster/variables.tf still said the Suramadu server and client JVM had never been measured together; envs/lance.blue/variables.tf and docs/architecture.md now record a 3.2GB peak, and terraform.tfvars sets match_task_memory to 8192. Only the description changes, not the default. envs/bootstrap/outputs.tf's init_command described a partial-backend init this repo does not use. backend.tf writes the bucket and region out in full, so plain 'tofu -chdir=envs/lance.blue init' is the real command, matching docs/runbook.md step 4.