I get it, it's not ideal. Yet there is no one else doing this work so I put effort where I think it's most neglected - which isn't fully refined prose, imo.
I call it LawVM - interpret amendment statutes as "programs" and verifiably replay them to obtain actual consolidated i.e. current law as it stands (or where ambiguous, surface that)
I was annoyed that in Finland there is no way to know what the law even says, it's basically a do-it-yourself endeavour and the "official" consolidated law isn't even official.
If the manual compilation/consolidation has any errors, then you're out of luck. Courts only decide based on the original statutes. And I have found hundreds of errors when doing the compilation.
This could've been done for >30 years and no one ever did.
Full release soon enough once I've cleaned it up. It's a whole compiler suite with Finland, Estonia, UK, Sweden, Norway to start with.
Part of a larger project to build the "state causal map" and doing AI-assisted analysis of all the mechanisms that comprise a state and therefore what is most harmful and what is optimal for governance. LawVM itself doesn't use AI at all except for development.
There is a simple way to mitigate prompt injection. Just check metadata only: is this action by the LLM suspicious given trusted metadata, blanking out the data
Interesting. I've been just now exploring something related or tangential, but for Finnish governance (why not all countries eventually).
Idea: Pull in all related docs and statements from the law and its process. And analyze its mechanisms across all timeframes, and across all capital stocks including social/human/moral etc capital. (finding: no one understands game theory nor causal models apparently)
I revised the essay, my priors were off. Added Good Samaritan data (zero successful CPR lawsuits in 30 years), duty-to-rescue statutes across Europe, and a new section on self-defense showing the trap cuts across Common Law/Civil Law (UK restrictive, Germany/Poland permissive).