Truth Surfaces in Legal Information Systems

When someone asks "what does the law say?", the answer depends on which surface they are reading.

Four surfaces

Enacted-event truth. What parliament actually decided. The original statute and every amending act, as published in the official gazette. In Finland, this is Säädöskokoelma. This is the legally primary surface — everything else is derived from it.

Published artifact truth. The machine-readable representation of enacted text. In Finland, this is the AKN XML published through Finlex open data. It should be identical to the enacted text, but sometimes it is not — XML production pipelines introduce structural errors, encoding artifacts, and editorial corrections that diverge from the gazette.

Revised/editorial truth. The consolidated text as maintained by an editorial team. In Finland, this is the Finlex "ajantasainen lainsäädäntö" — an informational consolidation with no legal authority. In Estonia, it is the Riigi Teataja consolidation, which is legally authoritative. In the UK, it is the legislation.gov.uk revised text, which is official. The legal authority of the consolidated surface varies by jurisdiction.

Replayed truth. The point-in-time text produced by replaying amendment operations from primary sources. This is what LawVM produces. It is derived independently from the enacted-event surface, without consulting the editorial consolidation.

Why the distinction matters

When LawVM's replayed text diverges from Finlex's editorial consolidation, the divergence is not automatically an error in either direction. It could be:

• A replay defect — LawVM's parser or compiler produced incorrect state.
• An editorial divergence — Finlex editors made a consolidation choice that differs from strict replay (e.g. adding repeal stubs, reordering content, applying future-effective amendments early).
• An oracle error — Finlex simply never applied an amendment that was enacted and published. The official consolidation is stale.
• A source pathology — the published XML is itself defective. Neither system can produce correct output from broken input.
• A corrigendum gap — a published correction exists but was not applied by one or both systems.

If you collapse all of these into "LawVM disagrees with Finlex, therefore LawVM is wrong," you miss the most interesting findings: the cases where primary sources support a candidate issue in the official consolidation surface.

Jurisdiction matters

The framing "official portals are editorial artifacts, not law" is too coarse. It is true for some jurisdictions and false for others:

• EU: EUR-Lex consolidated texts have "no legal effect" and are documentation only. The authentic version is in the Official Journal.
• Finland: Finlex consolidated texts are informational. The legally relevant sources are the original acts in Säädöskokoelma.
• Estonia: Riigi Teataja consolidated texts are legally authoritative by statute. If there is a conflict between the amendment chain and the consolidation, the consolidation wins.
• United Kingdom: legislation.gov.uk publishes "official revised" legislation maintained by the National Archives.

LawVM's role changes by jurisdiction. In Finland, it is an independent oracle that finds errors in the editorial consolidation. In Estonia, it is a consistency verifier for binding law — divergences are legally significant. In the UK, it is independent verification of the official version graph.

The operational consequence

A replay system that does not distinguish truth surfaces will either:

• treat the editorial consolidation as ground truth and silently hide its own errors and the oracle's errors alike, or
• treat its own output as ground truth and ignore legitimate editorial choices.

LawVM takes neither position. It compares surfaces and classifies the divergence. A mismatch is not one thing — it is a typed event with a root cause, and the root cause determines who is right.

This is why the Finland showcase reports both similarity metrics and a residual taxonomy. The number says how close the surfaces are. The taxonomy says why they are not closer.