An independent study by researchers with no connection to PiLogic or UCLA ran 18 methods across 672 benchmark problems. Ace, the academic engine built on the 2008 method, solved the most of any of them, and returned exact answers rather than estimates.
The partition function is the normalising quantity at the centre of most probabilistic reasoning tasks, and computing it is hard in the formal sense. Researchers at IIT Kanpur and the National University of Singapore surveyed 18 methods for it, exact and approximate, and ran all of them on the same 672 benchmark problems drawn from eight problem families.
By the study's own count, Ace, the academic engine that implements the 2008 method, solved 611 of the 672, more than any other method, and did so with exact answers rather than estimates. The authors wrote that among the exact methods Ace "should be preferred for exact inference." Their broader finding surprised them: the exact techniques were as efficient as the approximate ones.
None of the authors are affiliated with PiLogic or UCLA. It is the neutral test of the approach PiLogic's engine is built on, run 13 years after the method was published, by people with no stake in the result.
Ace is the academic engine, not PiLogic's commercial engine. Ace was not the best method on every problem family; on one, object detection, another tool matched the best result. The study measured partition functions, one task among several the engine performs.