It was because of two bugs: an ffmpeg bug (signalling chroma positions wrong) and a bug in their testing tool (which resampled the image, something that a PSNR tester should never do for any reason).

You're clearly wrong if you're implying they didn't retract that claim, as they do so here, in an update to the original page:

Interesting, last time I looked they hadn't, but that's quite welcome then. But of course this pattern has repeated itself a dozen times already.