Internal deployment, dated 2026-09-02. Source: F4304_MMC_CLIPS.md.
We ran MakeMyContent's own clip pipeline ("channel → clips → auto-post", shadow mode
— nothing was published) against an 84.4-second video we own outright. In 43 seconds of wall
time, the pipeline selected and rendered 3 captioned vertical clips (15.0s, 22.0s and 22.0s),
each with burned-in captions generated from real word-level transcription timestamps, a thumbnail,
and a drafted title — using six transparent, disclosed scoring features (speech density,
hook-cue phrases, emphasis keywords, scene-cut proximity, question-mark-in-opening bonus,
duration fit). No clip was published anywhere; every draft was written to our own database with
status draft_shadow for human review, exactly how the product behaves for customers
today.