A small side project with a large switch matrix
This project is one of freezy’s side projects. It began with a practical observation: VPE can offer table authors much better tools when it knows what is connected to each switch, lamp and coil. The premise is small; the number of machines is less so.
Better data, better authoring tools
A trustworthy machine definition will let VPE resolve ROMs, populate managers, map switches, lamps and coils, understand mechanisms, and suggest spatial assignments. The aim is to save table authors from repeatedly transcribing information that can be shared once.
AI handles much of the repetitive work
Most of this catalog is created and curated with AI agents. Repetitive extraction, large comparisons, address-by-address bookkeeping and cross-referencing are a good fit for them. They can work through the long lists; people can concentrate on judgment and the awkward exceptions.
Evidence before confidence
AI can also be confidently wrong. To keep hallucinations from becoming catalog entries, facts are tied to manuals, PinMAME source, known-working VPX scripts or runtime observations. Unknowns stay unknown, disagreements remain visible, and claims carry enough evidence for someone else to check them.
Spare inference has a useful home
Anyone with spare inference is invited to help make the database more complete. Pick a stub or partial machine, point an agent at the contribution workflow, require sources for its claims, and send the result as a pull request. The repetitive work can be delegated; contributors and reviewers still make the final calls.