It's a pretty click-baity title, it should be hedged to 'native floating-point' or something. In the second section the author clearly shoots down the title, w

It's a pretty click-baity title, it should be hedged to 'native floating-point' or something. In the second section the author clearly shoots down the title, while again choosing to repeat that it 'isn't enough', having again as a subheading: >Seven Bits Looks Like Enough But It Isn’t When that section clearly demonstrates that it would be enough. This statement really should be hedged.. You can just straight-up do binary-encoded decimal and change every pair of decimal digits into 7 bits and back again (bin 0000000 for decimal 00 through bin 1100011 for decimal 99, see reference [1] for chart), end of story and never suffer any degradation. I realize this isn't the scheme that is chosen, but that is a choice. Both the title and the quoted sub-heading imply otherwise; i.e. that it isn't enough bits. As shown above clearly it is. [[[ Aside: in this scheme you EVEN have some kind of 28% larger space you can make use of somehow for checksumming or the like! Naive example: when encoding any value that happens to contain digits smaller than 28 you can also encode parity information: when encoding, if the two digits you're encoding happen to be digits 00-27 you ALSO encode the checksum value 1 by choosing to add decimal 100 (to arrive at bin 1100100 through 1111111 i.e. decimal 100 through decimal 127) or checksum value 0 by passing on doing so (i.e. keep bin 0000000 through 0011011 i.e. decimal 00 through 27). When decoding, when you decode and encounter decimal 100 through decimal 127, subtract 100 from it before recording, and also record a checksum 1; when encountering decimal 00 through decimal 27 record it but also record a checksum value 0. This means that in 28% of cases you gain 1 bit of checksumming information while perfectly encoding two binary digits. (If your stream doesn't happen to contain any decimal digits 00 through 27 then you just don't gain any checksumming information.) You can use these checksum bits however you like! You can't be guaranteed to have them, but you have a good chance of having them, starting with a 27% chance of having a parity bit in the case of even a single pair of digits! I'm not an expert on checksums, and the above is just a naive example, but the point is this: not ONLY is 7 bits enough to encode 2 decimal digits: it's even enough to add a parity bit in 27% of input cases. ]]] [1] http://ascii.cl/conversion.htm http://ascii.cl/conversion.htm