I'm just wondering if the Travelling salesman problem could be used as a PoW algorithm?
Sort of. If the salesman has to visit a small fixed number (e.g. 42) of cities in a huge random
bipartite graph, then Cuckoo Cycle
https://github.com/tromp/cuckoo fits the bill...