Knobs Aurora respects across every run — adjust without a deploy.
Changes here apply to the next ideation/research run. Cache TTL is 5 minutes — running jobs finish on the old value.
