# 86TO88

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Chad Doebelin](https://cerebralvalley.ai/u/Astrobleem)
- **GitHub:** https://github.com/astrobleem/86TO88
- **Demo video:** https://youtu.be/L1UxVt3oN9A
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/244

86TO88 – the WM88 Bridge

The original, unmodified Windows 3.1 Minesweeper, running on a real 8088 Tandy 1000 EX under Windows 3.0. A DOS bridge built by Meta’s Muse Wonder translates the program in memory and never changes the file on disk.

Try it: https://astrobleem.github.io/86TO88/ (one click, desktop browser; emulated 286, not an 8088)
Video: https://youtu.be/L1UxVt3oN9A
Docker: docker build -t 86to88 . && docker run --rm -p 8080:80 86to88

The problem
Windows 3.0 refuses the file for its version stamp. The 8088 can’t execute hundreds of its instructions. The program also imports a library Windows 3.0 doesn’t have.

The bridge
A small DOS TSR (WM88TSR.COM) and a 361-byte SHELL.DLL fix all three outside the file. The TSR watches reads of that one known binary and translates the bytes in the read buffer only:

Version stamp: “requires 3.10” becomes 3.00 in memory.
Missing instructions: 449 sites using 80186+ instructions, plus 6 where a relocation overlaps a push immediate, become software-interrupt calls. A resident handler performs each operation itself and reproduces the exact result, flags and stack effect.
Missing library: the DLL supplies the one call Windows 3.0 lacks, ShellAbout.

Before translating anything, the TSR checks the file’s exact size and CRC-32, so any other file is left alone.

How it was built
Muse Wonder built the bridge, the test suites and the original validation. It started from a prompt that described the problem and pointed to where the public Windows 3.0 and 3.1 disk images could be found. Claude later built the hosted page, the video and the README, and did not write the bridge logic.

Evidence, kept separate

Real hardware: the original game launches and plays on a physical Tandy 1000 EX (8088, expanded to 640 KB), including numbered reveals and a loss. The hardware run used a hardened derivative of the published TSR. Photos and video are in the repo.
Independent emulator check: the published TSR in DOSBox-X launched, flagged, flood-revealed and exited cleanly. Without the bridge, the same setup stopped at the version message.
Development validation: under an emulated 286, a full game was solved and the game recorded its own best time.

Limitations

It targets one verified binary build. Other builds are refused.
It’s slow on the Tandy. I haven’t measured it, and I claim no hardware win and no benchmark.
The About box is a plain message box, since that is the one shimmed call.
Everything else, Windows and all game logic, is the genuine original code.

Next: per-program translation profiles, an option-ROM build that intercepts at the disk level, and real-machine performance measurement.

Repo: https://github.com/astrobleem/86TO88. It has source, builders, tests and evidence, and contains no Microsoft binaries.

## More from Meta Global AI Developer Hackathon

- [Watch Me](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/241)
- [VERA](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/242)
- [The Dealer Social](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/243)
- [Nguyen Quoc Anh](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/245)
- [SheetProof](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/246)
- [ReelDirector](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/247)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/244. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt