Decimal Exponent Floating like in JOSS

On my web page, at

I have a link to a document at

which is a 142-page .PDF in two columns. It is derived from what was a portion of my web page; I eliminated that for the time being for space reasons, but brought it back in this form as I found I could barely squeeze it in.

This is the description of an imaginary computer architecture; it is written to illustrate various computer architectural concepts by including virtually everything but the kitchen sink.

In any event, I noted that since I went as far as to include a compressed decimal feature (based on Chen-Ho encoding rather than Densely-Packed Decimal, since I use ten's complement by default... although on another page,

I note that DPD can be modified to handle ten's complement quanbreasties), I should also go even further, particularly as decimal floating-point is currently being seriously proposed, to include binary floating-point with a decimal exponent, as John von Neumann included with his JOSS interpreter (historically important for inspiring FOCAL, which used conventional binary floating-point, though).

Anyhow, if one wishes to make maximum use of the binary range, one has a small range of numbers that get an extra decimal digit of precision; that range is extended a tiny bit upwards part of the time if the test for "divide by ten" is binary carry out, and some denormal quanbreasties are allowed if the test for "multiply by ten" is if a fraction is less than 3-32nds - thus, one has an extra digit that is sometimes there, and sometimes not.

Although one has exact results for decimal quanbreasties within the stated precision, therefore, the behavior of such numbers in other respects is such as to horrify numerical analysts, I'm afraid.

John Savard