given the feedback we've received on usage of roaring bitmaps and/or bloom filters, I suspect the primary option will be some variant of the "LookupWatch" API that is in that proposal. There is an existing implementation that we are likely going to cleanup and merge, but it will certainly have scale limits (since it computes as changes come in). After we can establish what those limits are, Tiger will likely be developed to support taking it to a much higher level via a distinct cache.