Use one active brain per lane: the journal never picks, ranks, or switches providers on its own. Resolution is now explicit provider -> key-presence order (google/anthropic/openai) -> local-if-ready -> no-brain, with one default model per provider. The context string no longer influences resolution. Delete automatic provider fallback, model tiers, per-context routing, and per-interface backups. An active-brain failure now fails honestly with its reason code, recording a quota row from both executors first. Leave legacy tier/backup/contexts/models config keys inert and byte-intact on disk; no migration rewrites them. Fix a live local-lane bug where timeline rollup pinned a cloud model and failed on local installs.