Regarding performance, then i did a local test, all on same machine, so no network involved, using vanilla postgres, on a 5 mill relations data set. I got the same performace regardless of 100.000 or 5 mill entries in spiceDB, so a runtime complexity almost at O(k), which is very good. My particulars was 6ms pr write relation and 0,5 ms pr random read all across the data set. I noticed that the first 20.000 continuous writes was/is halv the execution time at 3ms pr write, is seem that is somehow buffers write to a certain extend then adds some processing time after that.