image and video scoring differently

[VD-NVencC]

I'm using the same material but getting different scores

take a lossless source, throw it through ffmpeg and notice image & video score differently
ffmpeg -hide_banner -colorspace bt709 -color_primaries bt709 -color_trc iec61966-2-1 -stream_loop 0 -i rgb.bmp -vf zscale=rangein=full:range=limited:matrix=709,format=yuv444p,zscale=rangein=limited:range=full:matrix=709,format=rgb24 -vcodec png y444.png
ffmpeg -hide_banner -colorspace bt709 -color_primaries bt709 -color_trc iec61966-2-1 -stream_loop 4 -i rgb.bmp -vf zscale=rangein=full:range=limited:matrix=709,format=yuv444p,zscale=rangein=limited:range=full:matrix=709,format=rgb24 -vcodec ffvhuff y444.mov

[edit]
source I'm using (rgb.png)

some history
followed a link from wiki.x266.mov/docs/metrics/SSIMULACRA2 which talked about _rs version
found myself following _bins instructions for the windows env github.com/rust-av/ssimulacra2_bin
I didn't follow _bins instructions precisely since their url 404's (tag latest of L-SMASH)
I supplemented AkarinVS's L-SMASH for Asd-g's https://github.com/HomeOfAviSynthPlusEvolution/L-SMASH-Works/releases/tag/1194.0.0.0

some newer history
I thought I should report errors to _rs github since the cargo called itself ssimulacra2_rs but the ssimulacra2_rs github page says it's version is only 5.1 and its crates io page is not the _rs version.. so I think this bug should be on _bin
I have since revisited AkarinVS's page and got their L-SMASH, but when swapped and doing this video test I get Python exception: lsmas: bgr0 is not supported.
I changed the ffmpeg cmd pixel format from rgb24 (reads back as bgr0 in ffmpeg) to gbrp for ffvhuff and the comparison succeeds, however both L-SMASH libraries score the same way.

for y420, I may not be apples to apples yet with chroma upsampling on (ffmpeg vs avisynth), meanwhile y444 scores exactly the same between them.