Claims of widespread errors in research involving Saez and Zucman

The audit site says it used Claude Opus 5.5 and OpenAI's GPT-6 Astra to check each paper separately, then had a third agent re-check flagged problems. It claims 69% of the 138 papers had at least one problem and 147 findings were serious enough to change a headline number. Tyler Cowen, quoted in a post, called for a “radical revaluation” of the work and associated policy arguments.