An actual TrollCPU

https://en.wikipedia.org/wiki/NCR_315

The instructions could use a memory slab as either two 6-bit alphanumeric characters or as three 4-bit BCD characters
The addressable unit of memory on the NCR 315 series is a "slab", short for "syllable", consisting of 12 data bits and a parity bit
A slab may contain three digits (with at sign, comma, space, ampersand, point, and minus treated as digits) or two alphabetic characters of six bits each.
A slab may contain a decimal value from -99 to +999.
A numeric value contains up to eight slabs. If the value is negative then the minus sign is the leftmost digit of this row.
There are instructions to transform digits to or from alphanumeric characters.

>>17
1.Normies and compiler writers BTFO(porting to ternary will require human labor)
2.Programmers salaries increase
3.Ternary provides best(closest to e(2.71)) information density
4.Neat bithacks exploiting middle-endian format
5.Ternary is faster and allow storing more data in memory.
6.AI switches to fuzzy ternary logic(true,unknown,false)
7.Networks which use Middle-endian ternary would be much harder to hack or eavesdrop into

>>18 A slight suggestion:
Protocol buffers encoding on top of middle-endian ternary.
https://developers.google.com/protocol-buffers/docs/encoding

Protocol buffers have many advantages over XML for serializing structured data. Protocol buffers:

are simpler
are 3 to 10 times smaller
are 20 to 100 times faster
are less ambiguous
generate data access classes that are easier to use programmatically
https://developers.google.com/protocol-buffers/docs/overview

Also Floating Point should use base 2.71(log e)
Its has many mathematical advantages and is optimal for data storage.
https://en.wikipedia.org/wiki/Natural_logarithm