@barrbrain I also saw your other commit (barrbrain@5b45ccf)

This was a mistake, I shouldn't have changed the scale.

I tried another experiment which gave modest gains on top of the rounding change, barrbrain@7baa93f :
"Do not round in get_mv_rate except for original resolution"

PSNR Y	PSNR Cb	PSNR Cr	CIEDE2000	SSIM	MS-SSIM	PSNR-HVS Y	PSNR-HVS Cb	PSNR-HVS Cr	PSNR-HVS	VMAF	VMAF-NEG
-0.0248	0.1399	0.0763	-0.0095	-0.0751	-0.0429	-0.0080	0.1179	0.1972	-0.0085	-0.1348	-0.1132

However, the more surprising result was that if that change were applied first, the rounding change had no impact on decisions.

Compared to the baseline, results for this experiment were a wash:

PSNR Y	PSNR Cb	PSNR Cr	CIEDE2000	SSIM	MS-SSIM	PSNR-HVS Y	PSNR-HVS Cb	PSNR-HVS Cr	PSNR-HVS	VMAF	VMAF-NEG
0.0010	0.1568	0.0512	0.0156	-0.0321	-0.0008	0.0159	0.0657	0.1465	0.0116	-0.0314	-0.0091