<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Armand Graaff — Notes</title><description>Notes on building and running AI systems in production.</description><link>https://armandgraaff.com</link><language>en</language><item><title>Measuring retrieval honestly</title><link>https://armandgraaff.com/notes/measuring-retrieval-honestly</link><guid isPermaLink="true">https://armandgraaff.com/notes/measuring-retrieval-honestly</guid><description>Most retrieval benchmarks measure whether the right document was returned. Almost none measure whether you would have noticed if it was not.</description><pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate><category>retrieval</category><category>evaluation</category></item></channel></rss>