Nostr notes by Steve Wetherill

I'm just going to leave this here; this is an NES version of ...

2026-09-08T03:02:23Z

In reply to nevent1q…586w
_________________________

I'm just going to leave this here; this is an NES version of Nodes.

https://claude.ai/code/artifact/5b2904a1-71c3-4692-8f2f-0ef581f19d9c

It's not perfect, but it's a pretty interesting proof of concept.

ZX Spectrum tape image -> playable NES version, courtesy of Claude Code.

#GenerativeAI #RetroGaming #GameDevelopment #ZXSpectrum #NES

Holy cow, you guys. I didn't think I'd be an AI ...

2026-09-07T07:35:52Z

In reply to nevent1q…dqhm
_________________________

Holy cow, you guys.

I didn't think I'd be an AI influencer, and I'm not, but I had to share this.

I posted earlier that I'd completed the BSS pass on the ZX Spectrum game Nodes of Yesod so I could add code for bug fixes, etc. Well, I was idly wondering what else I might do with it, and I've been rattling on about how it is all working, and blah, blah, blah, and it occurred to me that there might be a proof of life that I could share, at least temporarily.

What did I do? I wrote a "one-shot" prompt to convert the Nodes of Yesod Z80 assembler code (from the disassembly) to C and target WASM (WebAssembly). It took 42 minutes with zero input from me, and then I was playing Nodes of Yesod in the browser. I'd skipped sound for that first pass (I thought the beeper sound on the Speccy might be tough). In 20 minutes, the sound was working.

You can play it in your browser at the link below. This is NOT an emulator; it's Z80 assembler code converted to C and compiled to WASM. It thinks it is updating a Speccy display and getting Speccy keyboard input, and there's a light translation layer in and out.

So the cycle was Nodes of Yesod cassette tape image -> disassembled to Z80 assembler source code (and data) -> converted to C -> compiled to WASM. All with Claude Code.

https://claude.ai/code/artifact/0d0581c6-8f83-4514-b434-3cd46a903c06

I'll take this down at some point; play it while it is up. There are probably bugs.

#CodeArchaeology #RetroGaming #GameDevelopment #ClaudeCode #ZXSpectrum

I guess I'll just keep tagging onto this thread. The reverse ...

2026-09-06T18:59:56Z

In reply to nevent1q…hpfv
_________________________

I guess I'll just keep tagging onto this thread. The reverse engineering and disassembly of my game Projectyle are now complete on both Amiga and Atari ST. In each case, I started with a disk image. The 'BSS-ifying' step is still left to do so the source can truly be used going forward, but the semantic and symbolic disassembly is complete. Here's a nice screenshot of some source. :)

Additionally, here is a nice Claude artifact that plays the Atari ST Projectyle music. It is simulating the mixing done by the Quartet player on the ST, which bit-banged the YM audio chip volume registers to accomplish digital audio playback (I didn't write the Quartet player, to be clear).

It is a brave new world.

Here's the music player; this is the music I wrote for Projectyle in 1990-ish:

https://claude.ai/code/artifact/5aeabb7e-a0f0-4b6c-9e03-52b6ba8b6218

#CodeArchaeology #RetroGaming #GameDevelopment #GenerativeAI #AtariST #Amiga

Just in case anyone is interested in a peek under the hood, the ...

2026-09-06T06:05:40Z

In reply to nevent1q…vtcx
_________________________

Just in case anyone is interested in a peek under the hood, the changelist:

Nodes of Yesod — develop/ changelist

33 commits. Image 49,152 → 45,622 bytes. Persistent headroom 0 → 670.

Behaviour

- D1 — init_baddies overran its table. ld b,$00 cleared 256 bytes of a 128-byte table, writing $80 across $E200-$E27F — play_speech and 66 bytes of its sample. The first room change destroyed the speech routine; harmless only because that routine was spent by then. ld b,$80, same two bytes, no size cost. [003e175]

Memory map

- BSS left the image. $F236-$FFFF — back_screen tail, back_attribs, control blocks, mirror_table, stack, IM 2 landing pad — declared with rs, not emitted. −3,530 bytes. [c774974]
- baddy_state out of program space. 128 bytes of runtime state, first to $E380, finally into the slack above the player_sprites table at $B754. [35b7944, ea234d4]
- Sprite tables separated from sprite data. The two pointer tables need alignment; the 4,012 bytes behind them didn't. Costs 73, returns 128. [ea234d4]
- monolith_padding deleted — 103 zero bytes, no references. The only change that created space outright. [18dd094]
- speech_sample.bin — 4,246 db lines → one incbin. [e776706]
- 1,280 bytes of zero db → ds, byte-identical, ~40 interior labels preserved. [b955b85]
- astro_padding → ALIGN. Saves nothing — ALIGN re-adds it — but states the constraint instead of hiding it. [7744f4b]

If you have been following along, you'll know I claimed a ...

2026-09-06T05:55:00Z

In reply to nevent1q…krxu
_________________________

If you have been following along, you'll know I claimed a 100% disassembly of several ZX Spectrum games in my recent code archaeology endeavors. And it is true, BUT, the disassembly ONLY allowed me to recreate the original binary EXACTLY, byte for byte. If I added a byte of code to fix a bug, for example, it would break the build because everything would get pushed up in memory (and overflow by a byte in this case!)

Today I fixed that for Nodes of Yesod. Now, uninitialized data is no longer defined in the binary image. The Nodes code now uses BSS segments with RS statements plus alignment where necessary, and no longer forces everything into the DATA segment as DB statements.

The image shows the memory map changes. The speech area and the uninitialized area are both combined into RSS and used for runtime buffers.

As a result, today I fixed a bug that has been latent for 41 years!!! :)

It was a benign bug, but it let me prove the point by moving some buffers around in memory. Yes, the game still runs. :)

I plan on doing the same changes for the other 5 games.

#RetroGaming #CodeArchaeology #GameDevelopment #ZXSpectrum #GenerativeAI

More LLM code archaeology. I've hit a stopping point (edit: I ...

2026-09-05T07:00:01Z

In reply to nevent1q…jmsy
_________________________

More LLM code archaeology. I've hit a stopping point (edit: I mean they’re all complete) with the 6 ZX Speccy games, so I turned my attention to Projectyle, an Atari ST game I developed in 1990. Using the Claude Fable model, I tried a one-shot prompt to reverse-engineer the game and let it run overnight. It completed, and by the morning I had buildable source code, starting with a disk image as the only input. Here's a 'proof-of-life': a simple web viewer showing that the construction and rendering of the 8 arenas in the game are completely understood.

I'll post other examples in the coming days.

https://claude.ai/code/artifact/8c690c17-b070-4bf5-a8a0-b9f217a248ac

#CodeArchaeology #GenerativeAI #RetroGaming #GameDevelopment #Projectyle

I'm at v1.0 in my reverse-engineering efforts across Nodes of ...

2026-09-03T21:12:55Z

In reply to nevent1q…dqt4
_________________________

I'm at v1.0 in my reverse-engineering efforts across Nodes of Yesod, Arc of Yesod, Robin of the Wood, Heartland, Sidewize, and Crosswize, with each game 100% semantically and symbolically disassembled. The attached image shows some facts and figures from the resulting source code. In each case, bugs (or potential bugs) in the original games were discovered and documented.

#RetroGaming #CodeArcheology #GenerativeAI

Here's a little Claude code artifact, shared publicly, ...

2026-09-02T22:57:38Z

In reply to nevent1q…ymxf
_________________________

Here's a little Claude code artifact, shared publicly, showing progress on Crosswize disassembly. It's often useful to identify graphics data, name it, and then use the name to figure out render and move logic, etc. Here's a little utility Claude cooked up to let you name the sprites in the game. Never shared one of these before; LMK if it works/doesn't.

https://claude.ai/code/artifact/429a815b-6d76-4edc-ae4d-1d02c91c3140

For anyone keeping count, I now have Nodes, Arc and Robin at 100% ...

2026-08-31T14:57:52Z

In reply to nevent1q…m3pt
_________________________

For anyone keeping count, I now have Nodes, Arc and Robin at 100% semantically and symbolically disassembled. Robin is a particularly good test because it was done from scratch with no prior manual disassembly effort.

What do I mean by semantically and symbolically? I mean all code and data areas identified (table stakes), all symbols named (variable, functions), all major data structures identified and equates set up and used, so for example if there is a sprite control block structure with, among other things, x and y positions, then there are SCB_X and SCB_Y equates defined, and anywhere they are used would look like ld a,(ix+SCB_X), for example.

Additionally each function has a comment header describing entry, exit and preserved registers, a description of the function, and then within the function there are line level comments for things that might not be obvious (side effects, unconventional usage, etc).

All self modifying code sites are labeled, and commented with modifying sites.

Frankly, it’s in much better shape than the original source. I also have a list of defects for each game. 👀

Heartland is well underway and Sidewize has all code and data areas identified. Crosswize next, and there no rzx I can find for that so
I’ll need to play it through.

#CodeArcheology #RetroGaming #GenerativeAI

Onwards. :) Heartland. ...

2026-08-29T01:29:10Z

In reply to nevent1q…znku
_________________________

Onwards. :)

Heartland.

The other trick is to log the instruction boundaries from an .rzx ...

2026-08-28T13:14:49Z

In reply to nevent1q…4mzx
_________________________

The other trick is to log the instruction boundaries from an .rzx playback (ZEsarUX works well) and use those as instructions trace probes. Helps to identify code sections.

At this point I have working disassemblies of both Robin of the ...

2026-08-28T12:08:48Z

In reply to nevent1q…wvan
_________________________

At this point I have working disassemblies of both Robin of the Wood and Heartland. That might be a low bar, because technically 48KB of ‘define byte’ statements could be said to be a ‘working disassembly’ but it is a lot further along than that.

It’s quite fascinating how this works in practice. I go, “look for areas that look like graphics and render likely candidates as ascii art in the source”, then I’ll tell it, “those graphics are the main player sprites” and with that it can go figure out what code refers to those sprites, and infer the player control logic etc.

An interesting little detail is that, instead of using Ghidra, ...

2026-08-27T23:12:17Z

In reply to nevent1q…qe22
_________________________

An interesting little detail is that, instead of using Ghidra, Claude just wrote a Z80 disassembler. It is a little one-file Python script. And it uses it to probe unknown areas. Customizing as needed.

And just like that ... ...

2026-08-27T23:07:57Z

In reply to nevent1q…w27f
_________________________

And just like that ...

On to Robin of the Wood ... ...

2026-08-27T20:15:46Z

In reply to nevent1q…9y4g
_________________________

On to Robin of the Wood ...

I have Claude in a loop, just iteratively documenting and ...

2026-08-27T17:30:55Z

In reply to nevent1q…r477
_________________________

I have Claude in a loop, just iteratively documenting and improving the disassembly.

This is pretty wild. I did not type any of this. ...

2026-08-27T17:26:21Z

In reply to nevent1q…hlu2
_________________________

This is pretty wild. I did not type *any* of this.

Here's the Nodes video captured from Fused, built from my ...

2026-08-27T06:23:34Z

In reply to nevent1q…6ww3
_________________________

Here's the Nodes video captured from Fused, built from my disassembly.

Circling back to the disassembly of my 8-bit Spectrum games using ...

2026-08-27T06:21:58Z

Circling back to the disassembly of my 8-bit Spectrum games using generative AI.

The Nodes of Yesod disassembly is 100% semantically and symbolically complete and matches the original byte-for-byte. Attached source code screen and video show the results. I finally included Mark Butler's intro speech!

Today I started working on Arc of Yesod, and it now builds from source code with a 1:1 byte match with the original binary. I am now working on symbolicating it with a semantic review of the source.

I've posted on this before; the Nodes work benefited from a lot of manual work (by me) in Ghidra and IDA.

I only started on Arc today, and it clearly benefits from being a descendant of the Nodes source code, but I haven't done any hand disassembly. The toolchain is Claude Code, Ghidra (running headless via Claude), and numerous scripts Claude developed on the fly.

Once Arc is done, it is on to Robin, Heartland, Sidewize, and Crosswize.

#CodeArcheology #RetroGaming #GenerativeAI #ClaudeCode

(can't attach both a video and image, video in first comment)

so is it bayzil baggle? I am in the bazzle baygle camp, myself.

2025-01-21T23:39:16Z

In reply to nevent1q…js7g
_________________________

so is it bayzil baggle? I am in the bazzle baygle camp, myself.

Holy cow, I must have had some energy around the 2012 timeframe! ...

2024-12-22T18:43:51Z

Holy cow, I must have had some energy around the 2012 timeframe! I found I had not one but two versions of Nodes built from entirely different codebases, for Android! I’ll attach videos in replies since I can’t attach more than 1 here. The first one below is built in Flash using the Flashbuilder tool from Adobe, which could target native device builds. The second one is a ‘native’ (Java) Android build. Neither one was released (and neither is finished, but both are playable). I was planning to sell this device on eBay in an early spring cleaning purge, wonder if there’d be any interest if I’ll sell it with this ‘RARE’ unreleased software installed. #RARE #RetroGaming #NodesOfYesod #Android #Flash #FlashBuilder #Java (1/3)

The never-released version of Nodes of Yesod I made for Android. ...

2024-12-21T23:14:43Z

The never-released version of Nodes of Yesod I made for Android. Going through some old stuff, I found this Samsung Galaxy Player with a version of Nodes installed. I kinda forgot I made this. Excuse the dusty screen. :) #RetroGaming #Android #NodesOfYesod #GamesThatWereNot

The Spectrum from Retro Games Ltd is actually pretty good! It ...

2024-12-04T23:04:44Z

The Spectrum from Retro Games Ltd is actually pretty good! It comes with USB power and HDMI cord, and the rubber keyboard works great for playing games. Nodes of Yesod and Robin of the Wood are bundled, both play well. #RetroGames #TheSpectrum #ZXSpectrum #NodesOfYesod #RobinOfTheWood

this was great! Thanks for sharing. Ultimate were well ahead of ...

2024-11-16T20:05:06Z

In reply to nevent1q…qxjy
_________________________

this was great! Thanks for sharing. Ultimate were well ahead of their time with Knight Lore, it was visually stunning.

Makes my heart flutter. ❤️❤️❤️ It’s always ...

2024-09-28T20:39:44Z

Makes my heart flutter. ❤️❤️❤️

It’s always thrilling to receive nice feedback like this on games of yesteryear. :)

#RetroGames #Amiga #AtariST #Projectyle

https://blog.stevewetherill.com/2024/03/eldritch-cat-mk-iii-1990-1991.html

right. Never again, hard lesson learned.

2024-09-26T18:42:01Z

In reply to nevent1q…v200
_________________________

right. Never again, hard lesson learned.

I made this post on LinkedIn that seems to be getting a fair bit ...

2024-09-26T16:16:58Z

I made this post on LinkedIn that seems to be getting a fair bit of attention.

The post is about the fallout of installing the Microsoft Exchange profile on your iOS device, in my case so I could access company email & calendar in the native apps.

TLDR is i) don’t do this, because the Exchange admins can wipe your device (which is what happened to me), and ii) make sure you backup your device.

https://www.linkedin.com/posts/stevewetherill_psa-if-you-install-your-companys-microsoft-activity-7242971287369744385-7TVD

Another Suno cover version. This time from my ST/Amiga game ...

2024-09-16T18:56:27Z

Another Suno cover version. This time from my ST/Amiga game Projectyle.

https://soundcloud.com/total-harmonic-distortion/projectyle-manic-moose

And here's the Suno cover:

https://suno.com/song/a71dfb30-0920-4466-91aa-60665f4baad8

IDK what it is doing about the 2 minute mark, but it is totally something I might try, lol!

Saw this today, and that Nodes Of Yesod and Robin of the Wood are ...

2024-08-28T03:04:00Z

Saw this today, and that Nodes Of Yesod and Robin of the Wood are bundled, which was news to me.

A heads-up would have been nice, though. And here I am advertising it for them.

https://retrogames.biz/products/thespectrum/

tong or tongue?

2024-08-10T19:50:48Z

In reply to nevent1q…9rkx
_________________________

tong or tongue?

I added a photo to my blog post covering 1992/3 and my time at ...

2024-08-08T16:55:29Z

I added a photo to my blog post covering 1992/3 and my time at Electronic Arts UK. I just unearthed this one. I'm taking a break from debugging the game Risky Woods for the Sega Genesis/Megadrive using the SNASM dev kit from Cross Products.

#retro #retrogaming #retrogames #segagenesis #segamegadrive #devkit #electronicarts #riskywoods

https://blog.stevewetherill.com/2024/04/electronic-arts-uk-1992-1993.html

In 1993 I visited Las Vegas and interviewed at Westwood Studios. ...

2024-08-06T06:37:46Z

In 1993 I visited Las Vegas and interviewed at Westwood Studios. A new blog post detailing my experiences!

#WestwoodStudios #CommandAndConquer #LasVegas #Barnsley #ElectronicArts

https://blog.stevewetherill.com/2024/08/westwood-studios-part-1-1993-barnsley.html?m=1