Replay Bugs vs Editorial Artifacts

When LawVM’s replayed text diverges from Finlex’s updated/derived comparison surface, the divergence needs classification before either surface is treated as the explanation. Some cases are replay defects; others arise from source pathology, editorial treatment, or a candidate issue in the comparison surface.

A replay bug

A replay bug is a defect in LawVM's parser or compiler. The source material is sufficient, but the system produced incorrect state.

Example pattern: a coordinated repeal misread as section-level targets. An amendment can name several bare numbers followed by a single structural label. If the parser attaches the label to only the first target, it can replay a chapter-level repeal as several section-level operations. The source is sufficient and the defect is in target parsing, so the correction belongs in the frontend and should be measured in a fresh frozen run.

An editorial artifact

An editorial artifact is a divergence caused by editorial choices in the consolidation process — choices that can be defensible but differ from strict replay.

Example pattern: a future-effective amendment already present in a consolidated view. A comparison surface may include an amendment before the declared point-in-time query because publication and legal-effect dates serve different workflows. A replay that applies the operation only at its stated effective date is not automatically inconsistent; the record must state which date and comparison role were used.

A candidate comparison-surface issue

Sometimes the primary sources appear to support LawVM over the Finlex updated/derived comparison surface. Those cases are candidate findings until confirmed by Finlex or another competent authority.

Example pattern: a source-backed provision absent from the comparison surface. If an enacted publication and an admitted replay operation support a provision that the inspected updated text does not contain, the result is a candidate finding. Public adjudication requires review of the source references, comparison date, and authority response before anyone calls it an omission.

Source pathology

Sometimes the source XML is itself broken. Broken input cannot by itself establish a correct result.

Example pattern: malformed source payload. An amendment artifact can contain a stray token or duplicated word in replacement text. LawVM’s source-backed replay preserves the supplied bytes while the updated comparison surface may differ. The evidence class is source pathology, not a claim that either surface is globally authoritative.

In source-pathology cases, replay can be faithful to the supplied bytes while the source itself is defective. The system documents the pathology rather than silently correcting it.

Typed Divergence Classes

A single similarity score collapses all of these into one number. Even a high score can contain:

• replay defects;
• editorial divergences;
• candidate comparison-surface issues; or
• a mixture of those classes.

Without typing, it is difficult to improve the system, assess the witness, or publish a defensible finding. With typing, each divergence gets an explicit investigation path: a LawVM fix, a candidate comparison-surface issue, a source pathology, or an unresolved status.

This is why LawVM maintains a residual taxonomy and an internal finding record. In historical project analysis, 22 candidates that passed the project’s internal review threshold were reported for external review while many additional divergences remained under classification. The public site reports this aggregate only; the historical package and individual review packets are not published here.