Human-written amendment law is already executable.
LawVM compiles ordinary amending acts into point-in-time legal text-state — with typed operations, provenance, timelines, and explicit failure classification. No rewrite of legislation into a programming language required. The purpose is to prove the construction is possible at national-corpus scale, then make it easier for others to finish and institutionalize.
Reported findings
22
High-confidence meaningful replay-vs-Finlex candidate findings reported to Finlex for external review. Broader divergence evidence remains under active classification.
Divergence classes
15
Root cause categories. Replay defects, source gaps, editorial artifacts, and oracle divergences are explicitly separated — not collapsed into one error bucket.
Construction proof
0.65%
Mean text distance against the archived Finlex comparison surface in the frozen 2026-04-16 Finland benchmark snapshot. The point is not the score alone, but that replay gets this close from ordinary amendment acts.
Benchmark snapshot: 2026-04-16. Reported finding count: 2026-04-25. Corpus: Finnish alpha replay corpus curated from a larger amended-statute universe.
Research software. Not an official legal consolidation or legal advice. Finland frontend active; other frontends experimental.
amendment act → parse → typed ops → resolve targets → replay → PIT state → compare → classify
Deterministic by design: conventional grammars and symbolic replay, not text generation.
Why this exists
Legal text is not static. It is an incrementally maintained system: thousands of amendments applied over decades by distributed authors with partial authority. The law at any moment is the accumulated result.
The usual assumption is that executable law requires formalized drafting or post-hoc rules-as-code. LawVM shows a different path: compile the amendment stream already published by the legal system.
The immediate purpose is zero-to-one: prove that this can be done to a very high extent on a real legal system, not merely on toy examples or hand-authored rules. Finland is the proving ground because its amendment stream is difficult enough to matter and open enough to audit.
Most legal databases present an editorial consolidation — a human-curated snapshot. LawVM takes a different approach: replay the amendment chain from primary sources, preserve what happened, and make the result auditable.
That means typed operations instead of string patches. Point-in-time materialization from explicit timelines. Findings and source-pathology traces instead of silent cleanup. A way to separate replay defects from editorial artifacts.
LawVM v0.1 is therefore a construction proof and a handoff artifact. It exists to show that the substrate can be built, that ordinary law is already executable at the text-state layer, and that the remaining work is no longer speculative. Researchers, public institutions, legal publishers, and civic infrastructure teams should be able to pick up the mantle rather than restart from zero.
Read Why Law Is Law-Shaped for the full argument.
Carry this forward
LawVM v0.1 is a handoff point. The next stage should be pulled by people and organizations that want reliable legal text-state in real infrastructure.
Finland shows that the construction is technically possible. The next question is who will maintain the pieces that turn a proof substrate into durable infrastructure: reproducible benchmark releases, jurisdiction frontends, public-sector pilots, legal publisher audit workflows, and research replication.
If this matters to your jurisdiction or institution, the useful signal is serious adoption interest: a legal publisher audit workflow, a public-sector pilot, official source expertise, or a concrete plan to make LawVM part of legal information infrastructure. Code contribution by itself is not the bottleneck. See the handoff map.
Current state
Finland is the active reference frontend and the main proving ground. It is where the replay model, normalization rules, editorial adjudication, and residual-review workflow are exercised against a real corpus. See the Finland showcase.
Other frontends exist at different maturity levels: Estonia has strong consistency-checking telemetry, the UK is a substantial effects/version-graph workbench, and Norway/Sweden expose source-availability limits that shape what can honestly be replayed. See the jurisdiction status and benchmark caveats.
If the compiler model is still too abstract, start with the interactive explainer: it walks through a synthetic amendment from source text to typed operation to document-tree mutation. Open the replay explainer.
Layer 0
LawVM is deliberately narrow. It computes what the legal text says at a point in time. It does not compute what the law means (interpretation), how it is applied in practice, or what it costs. Those are higher layers that build on a correct text-state substrate.
The design principle: keep the text-state kernel narrow, explicit, stable, and anchor-rich. That is what makes it useful as infrastructure for everything above.
Get started
uv sync
uv run lawvm replay 2002/738 --as-of 2024-01-01
uv run lawvm diff 2002/738
uv run lawvm explain 2002/738