I don't think it takes doing anything wrong for diffuse spatial patterns to become well-defined when you substantially compress the space. What's dispersed across quite a few pixels in the full-size is now condensed into an easily discerned sharper edge of pixels.
If I defocus my vision, kind of like when looking at those old "MagicEye" 3D-like images, I can see the words pretty well in the full-size...
In the full-size there's a bunch of high-frequency noise (like building windows) interfering, but that's necessarily lost in the down-scaling, and now the low-frequency information forming the name is clearly visible.