Mapping priority nodes by response latency + cardinality is smart — most API audits I've seen just chase slow endpoints without asking "which ones does the checkout actually wait on synchronously?" The latter is what breaks revenue. Did you factor in cache hit rate per node? In some Indian ecom stacks I've touched, a seemingly "hot" node is actually served 90%+ from Redis, so the observable latency is hiding a cold-path cliff when cache is cold.