Potential answer readings expected, missing

In Turkish, some attitude reports alternate in veridicality with embedded declaratives. These, however, are uniformly veridical with embedded questions, but given a generalization due to Spector and Egré (2015), we expect them to alternate there as well. I present this puzzle of the missing potential answer reading and argue that two known restrictions on the distribution of embedded questions do not account for it, namely one based on non-veridicality simpliciter, and the other, on neg-raising