I built a product-video batch demo from the posting's specification block. Three synthetic placeholder sources stand in for raw footage (4K landscape, 1080p landscape, 1080×1920 vertical). An edit plan of 13 records, one per placeholder shoe style, takes 2 or 3 source ranges each (5 use three, 31 ranges in total) with a crop position. The build computes the largest 1080:1432 crop for each source shape (1628×2160 from 4K, 814×1080 from 1080p, which it flags as a 1.326x upscale, and 1080×1432 from vertical), cuts, crops and encodes all 13 files as H.264 MP4 with no audio stream, then reads each file back with ffprobe and parses the container: 13 of 13 are 7.50 s, 1080×1432, with zero audio streams and the index ahead of the media data; a bit-rate cap bounds any 7.5 s file at 4,156,250 bytes against the 5,000,000-byte limit, and the placeholder files measure 0.07 to 0.11 MB. Any failed check exits non-zero and writes nothing.