The npm packaging fix is the real lesson here, not the video pipeline. A files whitelist in package.json is a one-time cleanup until the test suite grows and a new benchmark render gets committed without anyone noticing, and the package creeps back toward 290MB. Worth turning npm pack --dry-run into a CI check with a hard size budget, so a regression gets caught before someone downloads a bloated CLI by accident. Separately, curious whether you've actually measured watch-time or click-through on a Reelcraft-produced reel against the old OBS recording, or whether the four-beat structure is still a design assumption without real before-and-after numbers behind it.