Native alpha delivery without visible artwork loss

A prospective in-memory PNG delivery transform clears only RGB hidden behind alpha0. It preserves every visible RGBA sample, every partial-alpha sample, all alpha values, native1254by1254dimensions and the original colour chunks. Catalogued originals and both selected native studies remain byte-for-byte unchanged. No runtime file is installed and no encoded test image or raw RGBA buffer is retained on disk.

For the ruined-wall study,the encoded PNG falls from3,127,867to679,031bytes. For the service lodge,it falls from3,382,458to85,440bytes. This is5,745,854fewer prospective transfer bytes across both images. Independent decoding checks the entire output. Normal browser HTMLImageElement and ImageBitmap paths each compare all1,572,516pixels per image with zero changed RGBA pixels and identical readback hashes. Source-bound measurements validation.json.

This proves a lossless visible delivery possibility for these two exact files. It does not reduce decoded memory:each still occupies6,290,064RGBAbytes. It proves no physical-device speed improvement,no normal placement/ground contact,no lifecycle acceptance and no source admission. Normal renderer integration must retain correct native coordinates and colour handling and repeat contextual,lifecycle and cost checks before activation. The core-preserved final artist PNGs remain the source of truth. Owned browser/server shutdown owned-resources-terminal.json.

A second complete filtered comparison covers both browser input paths at 0.125, 0.25, 0.5, 0.75, 1 and 1.25 scale, each with fractional translation, high-quality image smoothing and a coloured opaque underlay. All 24 full output comparisons are pixel exact. This closes that specific fractional-filtering concern for the two source-bound PNGs; normal world/lifecycle/backend and physical-device cost verification still remain. Filtered comparison filtered-validation.json.