This functionality was originally developed for Mongoid (MONGOID-3410), then extracted to the Ruby driver (RUBY-2130).

Having this capability at the drivers level can improve performance and reduce latency, while also providing another layer of availability for cached queries to be served during transient network issues (without having to wait for retryable reads to be performed).

Since only Ruby implements this functionality there is no spec, which would be needed to ensure a consistent implementation across languages.