They talk about this in the associated Numberphile video. The starting bottom-most card can be thought of as creating a partition of the deck into two parts: the ones "above" it are not uniformly shuffled but the ones "below" it must be, and so right at the moment when it is placed into the deck, the deck must become perfectly uniformly shuffled and any further shuffles cannot take it out of that state.
They also cover that overhand shuffles in practice require something like 10,000 shuffles to properly randomize. The problem is familiar to cryptographers: this scheme has "confusion" but not "diffusion". You can see this yourself: sort a deck of cards[1], then do two overhand shuffles in a row and splay out the cards and look at how random the result looks. They kind of "undid" each other, they "commuted" with each other or so.
If you want something a little more interesting, try to overhand shuffle with large cuts and then overhand shuffle with smaller cuts, and this gives somewhat more diffusion. Or just overhand-large, overhand-small, and then riffle -- the riffle will give you tremendous diffusion of the cut entropy created by the overhands. Anecdotally it seems somewhat unlikely that one gets to the full 220-something bits of randomness from only 7 riffles as that would require each riffle to have 30+ bits of entropy which seems... unlikely.
[1] you can do this fastest probably with a sort of omniscient quicksort: sort black/red, then sort black into spades/clubs, then sort spades into high/low, then sort the 6 low spades by eye, then sort the 7 high spades by eye, then sort clubs into high/low...
They also cover that overhand shuffles in practice require something like 10,000 shuffles to properly randomize. The problem is familiar to cryptographers: this scheme has "confusion" but not "diffusion". You can see this yourself: sort a deck of cards[1], then do two overhand shuffles in a row and splay out the cards and look at how random the result looks. They kind of "undid" each other, they "commuted" with each other or so.
If you want something a little more interesting, try to overhand shuffle with large cuts and then overhand shuffle with smaller cuts, and this gives somewhat more diffusion. Or just overhand-large, overhand-small, and then riffle -- the riffle will give you tremendous diffusion of the cut entropy created by the overhands. Anecdotally it seems somewhat unlikely that one gets to the full 220-something bits of randomness from only 7 riffles as that would require each riffle to have 30+ bits of entropy which seems... unlikely.
[1] you can do this fastest probably with a sort of omniscient quicksort: sort black/red, then sort black into spades/clubs, then sort spades into high/low, then sort the 6 low spades by eye, then sort the 7 high spades by eye, then sort clubs into high/low...