Unsurprisingly, there are more shenanigans lurking in TI-OS that the average user can stumble upon.

While within the VARS menu, some keypress that I cannot quite recall brought me to the Matrix editor, editing a matrix called "μ". I was able to set its size, entries, all that jazz, and then exit the editor. There is now a Matrix memory entry labeled "r²", which seems entirely inaccessible on-calc. TI-Connect CE calls this matrix "\", while CEmu refuses to even accept the file. The matrix appears normal on-calc, having the same size as a regular 2x2, though the file nabbed by TI-Connect is over 2 kb.

As to what keypresses got me there, I have no idea; I tried all single key combos from VARS. I would guess that it was something close to regular navigation presses, as they were haphazardly entered; it could not have been more than 3 subsequent keys.

The matrix itself appears unexceptional. Its name is the (unenterable) token 5C82, which appears as "r²" due to the way the token name strings are ordered in the OS, but putting it into a BASIC program would allow storing to & making use of this matrix. Similarly, it appears as "" on the computer side because the link protocol sends names as UTF-8, which gets decoded as "" from 0x5C & nothing from 0x82 (not being a valid UTF-8 sequence by itself). Normally it would encode the token into a UTF-8 string (so that 5C00 becomes "[A]"), but this fails here because 5C82 is not in the encoding table, & its default is to then just treat the token string as a UTF-8 string (presumably so program & list names work without needing a special case).

It grabs the wrong matrix when the computer requests the variable because it apparently does not check that the name is valid, so when "" fails to match any entry, it just returns the first one of the same type ([A] in this case).

On the other hand, whatever procedure produced it would be interesting to know. It is unfortunate that it was lost, but part of making accidental discoveries is sometimes not remembering how we got there. TI has similar bugs in the OSes on other calculators (even the TI-36X Pro, which is not a graphing calculator but has a visually similar menu system), but one problem is that sometimes they are context-dependent...so even the exact same keystroke sequence might not work without some other random condition holding.