What you are watching
The code holds a link to this site. Each notch of the slider flips more modules in the data area (finders are left alone, because if those go nothing helps). The decoder runs on every change and reports whether the link still comes out intact.
In practice
A logo is deliberate damage. If the visualiser shows level H surviving about 20% of flipped modules, that is the budget your logo, plus real-world wear, has to fit into. The logo safe-area tool applies exactly this test.
▸Reed–Solomon in one paragraph
The data is split into blocks. For each block the encoder computes extra codewords from a polynomial over a finite field, GF(256). Any combination of up to half the number of error codewords can be wrong and the decoder can still work out the originals, because the codewords are constrained to lie on a curve and a few good points pin the curve down. The deep dive has the arithmetic.
Try it yourself
Set the level to H, push the slider to where it fails, then open the logo tool and see how much of the middle you can cover.