And when the taxman takes exception to what Claude filed for you, you can get Claude to represent you in tax court. 

Claude and I had a chat about UP...I like that Claude suggested prompt improvements unprompted. It also understood why I moved the code before LOC_ADDR was initialized. It also displays a session memory for the NUMBER routine where I sidestepped this problem previously.
Does Claude know about DB ...

@BDD, I think you gave me the idea on how to access Forth's user pointer (UP), even after the direct bank is relocated. Basically, use an index register with absolute addressing to load the UP and the place it in the stack frame. Something like:

LDY #UP
LDA a:0,Y <——
STA LOC_UP
LDY #U_LATEST ...

I have to say, I don't quite see why it's so terrible not to have offset 0 available. Almost all 65xx programs work in a world where zero page (direct page) resources are limited, and most routines will have a limited range of addresses they can use. That 0 is such an unavailable address seems to ...

@BDD, I think I get it now. It's about aligning the expectation that the zeroth address of the local DP is within the local DP. However, if you only use named offsets you wouldn't notice this outside of the stack frame setup block. To change it I would need to adjust the offsets and introduce the ...

Update: I've concluded I'm confused about the INC before TCD. Since a subtraction pulls the stack pointer down, doesn't the INC push it up and increase the chance of overwriting the return address?
Not if you do things in the correct order. If, for example, you are allocating S_SF bytes for local ...

I think just had a rude awakening about changing the direct page register...the reason is that "STA 0,X" has the value of the DP register added to the zero offset.
Yep! All direct page accesses involve adding whatever is in DP to the address, even when DP is set to $0000 .

Something you should ...

* Claude must be paid by the word because he can be long winded.
It's been trained on MBA presentations. " Tell 'em what you're going to tell 'em, then tell 'em, then tell 'em what you told 'em. Three different ways ."
You can see that principle in action on Youtube, in which AI-generated ...

It's the nature of LLMs. It's not a person.
Yes, I know that. Despite what naive users might think of services such as ChatGPT and Claude, there is no sentience involved. It’s merely a stupid machine regurgitating information scraped from myriad websites.

I should have noted that my question was ...

After pasting the code, it fails to assemble due to illegal instructions...
Good catches. Those are both illegal on the 65816:
LDA (LOC_PTR,S) — stack-relative indirect requires a Y index. The valid form is LDA (LOC_PTR,S),Y, so Y must be set to 0 before using it.
I don’t get it. If Claude ...

After reviewing the changes, I am ready to render a verdict...While an interesting experiment, and I am amazed it works as well as it does. I don't see Claude as viable for this task. I certainly don't see it as worth paying for.
Makes me wonder if Claude would offer higher-quality results if you ...

Here are additional 65816 gotchas:

STZ only supports direct page and absolute addressing — no indirect modes. Use LDA #0 / STA (addr) instead
Nothing unique about that to the 65C816.

INC, DEC, LDY, LDX don't support stack-relative addressing — only LDA/STA do
In my experience, that hasn’t ...

There is this for the 65C816...not sure if it meets your desires.

Easy. All you need is 600nm fabrication factory :mrgreen:
Why stop at 600nm? Let’s take ’er down to 300nm and crank up the clock to new heights. Plasmo will be beside himself with joy at the thought of running his 65C02 at 60 MHz. :D

Speaking of plasmo, he hasn’t been here for nearly two months ...