Institutional assurance brief · 11 August 2026
Independent evidence for legislative change
LawVM reconstructs legal text through time from amendment sources and exposes the derivation for review. Institutions can begin with a bounded, read-only pilot beside their drafting, consolidation, or publication systems.
The institutional problem
A current text hides the transition history that produced it.
Legislation changes through source acts, effective dates, corrigenda, editorial consolidation, multilingual expressions, renumbering, transitional rules, and source-system migrations. A published text may be authoritative, editorial, evidentiary, or merely convenient depending on the jurisdiction. Comparing two final documents rarely explains why they differ.
LawVM treats legal text-state as a derivation. It preserves the source instruction, typed operation, target, temporal coordinate, resulting state, residuals, and comparison disposition so a reviewer can inspect the path instead of trusting a final string.
Project maturity
Beta-stage engineering with bounded assurance.
LawVM is open-source under the MIT license, with substantial working compiler, CLI, API, frontend, test, and evidence surfaces suitable for bounded evaluation. It remains pre-1.0: public interfaces and evidence formats can change, frontend maturity is uneven, and an exact result still depends on its declared software, sources, profile, corpus, and reviewer boundary.
The bounded assurance chain
Seven questions replace one “verified” badge.
Source identity
Bind the work to exact manifestations and roots. Identity does not prove source completeness or correctness.
Total accounting
Keep emitted operations, typed rejections, typed observations, and unaccounted effects visible inside the declared universe.
Semantic closure
Require typed target, payload, source anchor, and time. Preserve missing proof instead of guessing.
Mutation account
Compare writer receipts with independently observed before/after paths on wired profiles.
Temporal state
Separate effective time, expiry, temporary overlays, occupancy, identity, branch, and authority regime.
Review boundary
Expose what a separate checker or human reviewer can reconstruct and what still must be trusted.
Where institutions can start
Four read-only evaluation formats
| Question | Declared input | Return |
|---|---|---|
| Does a consolidation follow from amendments? | One source-complete transition set and comparison surface | Source census, replay account, classified differences, evidence packets |
| Do official-language expressions stay structurally aligned? | All declared expressions, versions, and a bilingual legal reviewer | Availability, pairing, invariant-fact, and residual account |
| What can be built from our archives? | Representative APIs, XML, HTML, PDFs, scans, registers, and access constraints | Source/page account, claim-specific reconstruction evidence, residual map, and transition feasibility |
| Can we check a transition before publication? | One versioned structured export and declared rules | Read-only target, reference, temporal, language, and mutation checks |
What a pilot returns
Evidence products for human review
Source and scope manifest
Authorities, manifestations, acquisition dates, hashes, missing sources, work/time boundary, and comparison roles.
Transition account
Typed operations, blocked candidates, recoveries, source pathologies, unexplained residuals, and unchecked units.
Human review queue
Exact source locators, before/after witnesses, confidence limits, alternative explanations, and disposition fields.
Promotion recommendation
Which checks can remain observational, which can block a shadow result, and what evidence is needed for the next claim.
Status discipline
Missing proof and contradiction lead to different decisions.
| Outcome | Institutional meaning |
|---|---|
| Established within scope | Every declared obligation for the exact claim and profile is satisfied. |
| Qualified | The result depends on named allowances or limitations. |
| Blocked | A required proof or authority input is missing; no promoted result follows. |
| Unresolved | More than one supported explanation remains. |
| Invalid | Integrity, trace, invariant, or accounting contradicts the claim. |
| Uncheckable | A required committed artifact is unavailable. |
| Not claimed | The property, source lane, family, or conclusion lies outside scope. |
Evidence already visible
One confirmed case, two public candidate packets
Riigi Teataja reviewed a LawVM-reported omission in Estonia's Auditors Activities Act and corrected the Estonian consolidation. The corresponding English-translation detail remains a project report pending separate adjudication. This one externally adjudicated case supports the case-specific QA claim.
Five additional source/current-text candidates reported from New Zealand and UK work are public with official source links. They remain candidates: publisher notification, acknowledgement, or a later public-text change is not confirmation.
Finland provides a dated replay-construction experiment over a declared corpus. Estonia exercises publisher consistency. The United Kingdom, New Zealand, Norway, Sweden, European Union, and United States exercise other source and effect models. Japan, South Korea, Poland, and Switzerland are staging lanes with explicit source-level claim ceilings and next gates.
Decision and next step
Begin with one source-complete question and a named reviewer.
Define the jurisdiction, institutional role, objective, representative sources, bounded corpus or transition family, and the person who can adjudicate local source authority or drafting practice. LawVM can then propose a shadow evaluation whose strongest possible conclusion is agreed before processing begins.
Contact: [email protected] · lawvm.org/pilots · github.com/eliask/lawvm