Thirty antitrust agencies report on their computational work, in what remains the only account written by the agencies themselves. The primary focus of this edition was one question: what these tools have allowed them to achieve that they could not have achieved otherwise. Some answer with a case they would not otherwise have opened. Others answer that what changed is the range of conduct they can examine at all.

The report gives a full account of the tools now in use. Three developments stand out against earlier editions. In the absence of a dedicated legal regime, some agencies have begun writing their own rules on what may be put into a model and what may not. Agencies are also putting figures on their work, reporting precision, recall, and false-positive rates, which earlier editions almost never did. And what holds deployment back is less technique than access to data and the agreements needed to obtain it. The contributions also show different strategies, including how much to build rather than buy, and where each agency places the line around what may enter a model.