According to the arXiv abstract for DuoDiscover, evaluation has been central to advances in AI and is equally critical to autonomous discovery. The authors argue that scaling autonomous discovery to real-world engineering challenges is currently blocked by a fundamental bottleneck: the cost of evaluating candidate solutions.

The abstract is truncated, so the available text does not describe how DuoDiscover addresses this bottleneck or what experiments were run. The paper's title suggests a method for scaling discovery beyond the evaluation limit, but the specifics are not present in the source.

Because only the abstract's opening lines are available, this summary is limited to the problem statement and the proposed system's name. No comparison with other approaches is possible from the given text.