My guess, based on sibyl’s work, is that it takes the whole message as a number, and then converts that number to a different base (possibly a base which varies with the place value). If this is true, then changing the value of the plaintext by 1 will change only the tail end of the spamtext, except in cases where it rolls over. Testing this:
abcde returns
abcdf returns
…and, so much for that. I can’t spot the difference immediately, but it’s definitely not in the last line.
EDIT: The difference is the punctuation after "This is different than anything else you’ve seen ".
The author of spammimic may be Peter Wayner, and this book might have some clues, but the 3rd edition seems to cost $60. I’ll see if I can make any sense out of the 2nd edition (2002), available for much less. The spammimic program was first noticed here in 2004, so maybe the newer info isn’t all that important.
Peter Wayner has responded to my email, and I will post what he said here if he gives permission to do so.
Meanwhile…to paraphrase, he is not the author of the program, but says his book(s) may have been the inspiration, and provide the technique(s). I am hoping he will contribute a little more than that.
Please note that there seems to be a new, non backwards compatible algorithm at play now. (See the decode page)
I had the idea to use emoji’s as plain, which works as expected. as input gave me some text, notably including “71 weeks”, which i changed to 72, giving me , as well as 73, giving
The original emoji is utf-8 hex 1F600, 72:1F610, 73:1F612.
I think if you get enough data you could see the binary/hex change in predictable patterns.
The support of emoji would support the idea that it is working on hex/binary level.
Since i cannot edit anymore: it is not linear, 69:1F606.
Any other change of numbers, i.e. the “Title Section Part”, results in single OR multi char representations of weird characters, the weeks part seems to affect the lower bit changes.
I realise that I maybe a few years late but I couldn’t help but read this, and so here I am to maybe make things a little clearer? (I barely know how to use this chat but I’ll try!!)
First of all, kellner was amazingly helpful with his idea of using binary codes from ASCII numbers, and was quite a useful role. However, Musicat pointed out that only 4 extra characters were added to a string with supposed 7 bit numbers! So what’s going on?
Each phrase in the text is given its own number, and the ASCII binary chooses the phrase given: this is why @ to underscore ALL start with “Dear Professional…” in some form, as the phrase represents the first 2 bits in their binary: 10. Extra starters are added on as well, along with , and ; meaning 0/1.
` to (DEL) change starting greeting every 8, as sibyl pointed out, so it scans the first 4: 11??
Every single line looks like it can be changed through binary, even the numbers, which I’m pretty sure are coded to have no mathematical meaning and instead are also hard coded: Have you ever noticed that the “section” never loses its starting “30”, or that there are only 8 different Senate bills?
So, at the moment, I’m looking at it like it could be a string of 1s and 0s. But there are some mistakes as well that could be noted:
-ASCII letters ending in 10 or 11 change the rest of the message entirely, even the stock at the end, but only in some cases, ending in 3 different stocks: some further binary trickery could be at work
-This also can affect the start of the message (e.g. “aa`" gives you an E-commerce prof. starting point instead of just “Professional”. Notice ` in binary is just 1100000 which the ending could be confused as stock (which is probably infinite 0s) and so needs to be changed in some way. The usual start of aa` is “~” or 1111110 meaning the text added a bunch of 1s..?
Everyone here is a great help and this is a fun investigation!!
wow this is an essay but I hope that this puzzle can still be solved after all of this!
as input gave me some text, notably including “71 weeks”, which i changed to 72, giving me 
, as well as 73, giving 