from bitcoin core

        consensus.powLimit = uint256S("00000000ffffffffffffffffffffffffffffffffffffffffffffffffffffffff");

We store this as (considering the floating point truncation)

0x00000000FFFF0000000000000000000000000000000000000000000000000000 

because the above value is not representable by compact target. While the result for both operations is the same and would result in the same block acceptance, I believe the former matches the core semantics better.

I believe this should also be considered a bug as it changes the output of difficulty function. Ideally, we should be able to cross check this operation with other exploerers/implementations.