La Frontière

First, a Python pipeline collects new publications three times a week (Mondays, Thursdays, and Saturdays) from arXiv (econ.EM, econ.GN, cs.LG, cs.CL, stat.ML, filtered for economics), the NBER, the journals of the American Economic Association (AER, JEP, JEL, AEJ), VoxEU/CEPR, the Bank of Canada, the U.S. Federal Reserve, the European Central Bank, and the International Monetary Fund. It then assigns a thematic score to each entry: the main selection includes those with a score of at least 3.

Next, for recognized economic journals and institutions (AER, JEP, JEL, AEJ, NBER, VoxEU/CEPR, Bank of Canada, U.S. Federal Reserve, European Central Bank, International Monetary Fund), the economic nature of the text is assumed: only the connection to artificial intelligence remains to be evaluated, so as not to exclude an article simply because its abstract lacks standard economic vocabulary.

Finally, the summaries are generated by a language model (Gemini, supplemented by Claude), then checked by automated validators: any figure absent from the title or the source abstract causes the text to be rejected. The English summaries are produced by the same pipeline; when one is missing for an entry, the page falls back to the French summary rather than leaving the card empty.

The signal of the week is the article with the highest thematic score among those reported in the most recent collection, not among the ninety days of the selection: without this freshness requirement, a highly rated article would keep the spot for weeks. A floor is also applied. When no recent article reaches a sufficient score, the section says so rather than highlighting a text that does not deserve it.

Two terms recur in the entry labels. A research paper is a work released before publication, a preprint or a working paper. A published article has appeared in a peer-reviewed journal or in an editorial publication.

I also review what the pipeline produces as far as possible. If you notice an anomaly or see room for improvement on any aspect, write to me: your feedback is welcome.