Or does FEND actually behave differently when it ends a FOR compared to when it ends a FMT?
"The Graphics Programming Language" by R. A. Green has this on page 81 "E.2 Format Suboperations"

FB FEND >FB {End of FMT}
FB FEND >FB,GGGG {End of FOR}

But on page 38 "9.29 FEND—End of Formatted Screen Write" it specifically says there is no operand:

[label] FEND [comment]

This instruction is a suboperation of the formatted screen write instruction and is only valid after
an FMT instruction. The FEND instruction terminates a FOR group if one is active, or the FMT
instruction. The instruction has no operands. STATUS is unaffected by this instruction.
Although a “label” is allowed, it is illegal to branch to a formatted write suboperation.
Example:

FMT Formatted screen write
··· Suboperands
FEND End of formatted write

And this GGGG doesn't occur anywhere else in the book. So now I'm not sure at all...

First of all, thanks alot for the bug report! It seems that the FOR argument is disassembled incorrectly. It's also a cool project you're doing!

Regarding the FOR-END, this weird address argument does exist, and GROM cartridges containing this statement do execute it. Of course, my disassembly is also wrong.

That's just my first glance, I'll have to look at it again later. Please note, however, I'm going on vacation later this week, and I don't know if I have time before that.

I'll keep my eye out for your fix to see if it matched RetroGhidra's output.
Let me know if you have a look at my project. I haven't pushed the GPL disassembler yet because I haven't worked out all of the addressing mode binary encodings yet.

First of all, thanks alot for the bug report! It seems that the FOR argument is disassembled incorrectly. It's also a cool project you're doing!

Regarding the FOR-END, this weird address argument does exist, and GROM cartridges containing this statement do execute it. Of course, my disassembly is also wrong.

I think I called this wrong. As my disassembly gets more complete I understand things better. It now seems to me that the same opcode is used for FEND to end both FMT and FOR but that it takes an operand only when it ends a FOR. Do you know if this is documented?

By the way, I'm also having some trouble decoding the operand format for MOVE. I know the regular operand format encoding is documented but unless I'm missing something the MOVE variant doesn't seem to be.