transcode-mcp — half-class overshoot, live

Each tile below requests the source image through the proxy at a different target width. The proxy computes the encode dimension via ceil_to_mod16(min(target × 1.5, source × 1.5)) and returns the transcoded image. The displayed dimensions and file size come from the X-Transcode-* response headers — nothing here is calculated client-side. Spec: encode-resolution-arithmetic.md

Loading source…
target × 1.5 binds (normal case) source × 1.5 binds (tiny-source cap) source equals target → click any tile to view at target size
Test matrix — target widths 
320, 480, 640, 720, 800, 854, 960, 1080, 1280, 1600, 1920, 2560
How to read the tile metadata 
source  — original dimensions (from X-Transcode-Source-W/H)
encode  — what the proxy actually rendered (from X-Transcode-Encode-W/H)
binds   — which term of min(target × 1.5, source × 1.5) bound the result
quality — numeric quality parameter passed to the encoder
format  — image format negotiated
size    — Content-Length of the response body
cache   — X-Transcode-Cache: HIT, MISS, or PASS (passthrough)