I think FIG-FORTHs UP is the beginning of supporting multiple users or tasks.
The User Area stores everything unique to a task.
The system could have multiple user areas. UP points to the current one.
The FORTH word PAUSE switches to the next task in a ring (cooperative multitasking).
The user area ...

Thank you for writing about your adventures with Claude! It's been interesting following along.

I assume Claude has no memory of previous sessions with you, except for the session logs you play back to it at the start of the next session. So to improve Claude's 65816 programming, one needs to play ...

There are some more word tests at https://github.com/leepivonka/FM/blob/main/test.fs

The Claude version of FORTH seems not know about colon words. It either ignores them, or converts them to machine code.

Maybe like:

u32_cmp: ; compare lba with (YA) as uint32, return P.C & P.Z
sta temp1 ; save ptr to arg
sty temp1+1
ldy #$ff ; start at most-significant byte
@1: iny ; step to next byte
cpy #4
beq @9 ; if done, return z=1, c=1
lda lba,y ; compare
cmp (temp1),y
beq @1
@9: rts

Another option is to use more traditional fixed 2 or 4 byte cell size for the primary
FORTH parameter stack & to define additional words & stack for the large integer cells.
Sort of like what ANSI FORTH does for floating point words.

Here is sample set of definitions for separate large integers ...

Did you mean to use A4 instead of A5 for CS1 on the 6522?

Do you need to write back to the interrupt status register to tell it you've serviced the T1 interrupt?

It seems like FIND is a compatibility carry-over from earlier FORTHs.

PARSE-NAME returns the found word in addr,u style that points into part of the input string, no copying needed.
It is easy to pass addr,u into something like FIND-NAME and the word header compilation word.

Assuming your serial port is implemented as a UART with at least 1 char of buffering, all you can say is your read file routines are on average at least as fast as your serial port.
While the serial port is processing the last TX char, the CPU can compute up to 1 char time (maybe 1 start bit, 8 data ...

.org is important! It tells the assembler where the following code is in the memory space.

If the assembler thinks the code you are JSRing to is at $802f it will code that absolute address into the instruction. When it executes, it'll jump to $802f & start executing data not intended to be ...

Are all of the VDD & VSS pins on the 6502 connected?
Is RDY confirmed always hi?
Is SYNC running?

How does phase 0 (clock input) of the 6502 look like when the clock outputs are not switching?
How do signals around the 74HC161 that generates PHI2 look?
My guess is an inadvertently disconnected CMOS input around the clock generator circuitry.

Many memory devices work as follows:
CE = chip enable = wake up (if this is inactive, OE & WE don't matter)
OE = output enable = turn on data bus drivers (often faster delay than CE)
WE = write enable = remember what is on the data bus (often faster delay than CE)

It is often faster to activate CE ...

Arg! I've replaced 65816S.exe with a newer version that does have the V command!

Source is at https://github.com/leepivonka/65816S

The 0265sxb.s28 link is fixed.

The emulator can load binary files with the V command. I use this to run Tali Forth:

V 8000 taliforth-65816s.bin
Z
G

V loads the binary file taliforth-65816s.bin at an offset of $8000.
Z does a reset, to pick up the Tali reset vector.
G starts execution.

If ...