Your choice of punctuation, most of which cannot be represented on a 7 segment display, is arbitrary[1], ASCII/US keyboard centred, and only added to your argument to increase bit count.
[1] why include /-+ but no *, =? # but no @, &, $, %? () but no []{}<>? " but no '?
Why /-+
Don't interpret them as math. Those are "or" (as in yes/no), hyphen, and "and" (as in you+me). I rejected ampersand because the use case was covered by +. I included one set of parenthesis because parenthesis are a gramatical necessity as scope can otherwise be ambiguous, but more than one style of parenthesis is redundant. I reluctantly added # because in order to fit everything in 7 sev, you are sometimes forced to use 1337 5p34k. So it might be necessary to have a symbol for "read this as a number". Contrary to your assertion that I was just padding out the count, I was actually trying to do the opposite and keep the extra symbol count down to only what is needed. You could cut all the special characters and still be at over 5 bits.
You said T and V twice. I already admitted elsewhere V,X,K are problematic. G -> g -> 9. Z -> 2. S being indistinguishable from 5 means you are forced to accept collision between numbers and letters anyway, so by the conventions established in 1337 5p34k, T -> 7 is an acceptable substitute. Accepting that there is only one form of a letter, no more capitals, you can do "(" with what used to be C, and ")" with a backwards "C". "/" is absolutely doable with 7 segments.
Of the 50 symbols I started with, there are just 9 that don't really fit in a visually intuitive way. KVX.,;+#
And honestly that's fine. The point of this exercise wasn't alphabet reform. The point was that 7 is a close estimate for the number of bits needed to define the shape of a letter in the English alphabet. The fact that the 7-segment system can cover just over 80% of these glyphs without anyone intentionally designing it that way is suggestive of the idea that alphabets naturally optimize towards some system of n-segments.
>You cherry pick between lower and upper case letters, discarding half of them, only to shoehorn them into your model.
Not really sure whats the problem here. I've been consistent. 1. The distinction between upper and lower will no longer exist. There is just one glyph per letter, sometimes the glyph formerly known as "upper", sometime the glyph formerly known as "lower". 2. Some glyphs that might have been used as an upper/lower version are instead used to encode something that otherwise wouldn't have a glyph (like m, w, parenthesis).
It's not shoehorning to repurpose the symbols I previously removed for redundancy.
>The way you proposed to represent M and W is non-intuitive
Try writing out a few sentences this way. I assure you it has enough in common with the outline of M and W that you won't trip over reading it.