I’ve thought about this on occasion. It seems that allowing arbitrary work units (eg user defined ones) breaks the POW model as it’d not use a predictable amoun

I’ve thought about this on occasion. It seems that allowing arbitrary work units (eg user defined ones) breaks the POW model as it’d not use a predictable amount of compute. Miners or others could readily game the system. Maybe it’d be possible to make the compute time unpredictable (or more predictable) using homomorphic encryption or similar techniques. Then a computation would operate over all the possible inputs. That’d be interesting, but would limit the types of problems. Alternatively, perhaps something like QM simulations over pre-specified inputs (with a salting function based on the current block hash) might do POW and be useable. The inputs could be just defined over an increasing number of atoms in set pattern. Though finding problems sets in QM that are also NP complete might limit the usefulness. Many interesting QM calculations aren’t NP and don’t have any polynomial way to check answers.