Claiming to solve an np hard problem in polynomial time on all inputs would be either a fraud or a breakthrough. This is not such a claim. The algorithm is organized to perform well on many but not all inputs--its worst case is exponential time and it doesn't pretend to be otherwise. If you play chess against a chess engine like Stockfish, the exact same thing is going on, and in fact the algorithms involved are closely related to the one in the article.
Claiming to solve an np hard problem in polynomial time on all inputs would be either a fraud or a breakthrough. This is not such a claim. The algorithm is organized to perform well on many but not all inputs--its worst case is exponential time and it doesn't pretend to be otherwise. If you play chess against a chess engine like Stockfish, the exact same thing is going on, and in fact the algorithms involved are closely related to the one in the article.