<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/c01032e2e58b40a19394404b8f78d5a8&quot; frameborder=&quot;0&quot; width=&quot;1720&quot; height=&quot;1290&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1290</height><width>1720</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1290</thumbnail_height><thumbnail_width>1720</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/c01032e2e58b40a19394404b8f78d5a8-fbb88442e6037d41.gif</thumbnail_url><duration>166.272</duration><title>ClickHouse tuning, live — same query, 1.8 GB → 6 MB scanned (runnable + CI-verified)</title><description>A 3-minute runnable proof of database tuning and CI regression prevention. Same 50M-row events table in two schemas — naïve vs tuned. Scanned data drops from 1.8 GB to 6 MB, and a GitHub Actions job fails the build if the tuning regresses.</description></oembed>