http://www.ibmcomposer.org/SelComposer/Manual/toc.htm
What I do find, in the appendix, is a list of characters and their respective relative widths. - an A is 8 units thick, a l is 3 units thick, etc. This is how it does porportinal spacing. I assume that, combined with how the letter is centered in its block, and how far the machine advances with that letter composes character metrics. Right? There's no mention in here of variable character metrics depending on the next character, which is what you would have to have to do what we are seeing, right?
Also on this page, http://pfaedit.sourceforge.net/editexample5.html which talks about character metrics, it says under kerning:
If you are careful in setting the left and right side-bearings you can design your font so that the spacing looks nice in almost all cases. But there are always some cases which confound simple solutions.
Consider "To" or "Av" here the standard choices are inappropriate. The "o" will look better if it can slide more to the left and snuggle under the top bar of the "T". This is called kerning, and it is used to control inter-character spacing on a pair-by-pair basis.
In the example I posted, the a snuggles under the bar of the t, which is exactly what he is saying.
Either way, the width of the at and ta pairs in Word and the memo are the same and there is overlap between the tail of the a and the bar of the t. In the example of text someone posted on an IBM Selectric Composer, which is the typewriter all the apologists say could have typed the memo, there isn't overlap. This is because the a doesn't have much of a tail, which is why the Press Roman Font has similar spacing to the Times New Roman font, because it doesn't have all of the serifs to deal with that TNR has.
Whatever you call it, it's there in the memos and word, and it's not on the text from the Selectric using the Press Roman Font where someone tried to copy the memo, and got close, but not close enough.