On a Heap Table, what does a non-clustered index use as a pointer to a row?

Pick your version of SQL Server... if it has changed between version please spec开发者_运维百科ify if you know.

Build a table Add a non-clustered index on 1 or more columns. If I could dump a leaf block, what would I find as the pointer to the row in the table with the data?

It uses a row id, which is basically a physical address of the row in the database: on Page X at Slot Y. If the row ever moves, then a Forwarding record is left in place so that the non-clustered index reference is kept valid. See Forwarding and forwarded records, and the back-pointer size.