I was very careful not to overstate WebP lossless quality. I compressed the originals with zopflipng and pngcrush before stating figures (-26 %).
If I had just used the internet quality images, WebP lossless would have improved size by -42 %.
Yet another 'marketing textbook' way to overstate -42 % is to turn it into 'loading speed': 1/(1-0.42) = 72 % faster.
None of this was made for the lossless and the most conservative estimate was shown. I didn't do any hacking or cherry picking to produce the number and I had several internal verification approaches to be correct.
If I had just used the internet quality images, WebP lossless would have improved size by -42 %.
Yet another 'marketing textbook' way to overstate -42 % is to turn it into 'loading speed': 1/(1-0.42) = 72 % faster.
None of this was made for the lossless and the most conservative estimate was shown. I didn't do any hacking or cherry picking to produce the number and I had several internal verification approaches to be correct.