There's something I don't understand about the cursor. Should it be opaque? Or should one look inside the base64 encoded data?

When I use a larger limit parameter (say 999 instead of the default 21), I get 999 sized chunks, but the cursor still 'steps' only 21, causing superfluous numbers of calls and duplicates in the results.

Only when I base64 decode the cursor and manually increase the offsets do I get the complete and unique set of results in a minimum number of calls. Is this intended?

Example: https://cl1p.net/sn