Artifacts

Methodology, metric, and reproduction boundary.

Benchmark methodology

LawVM benchmarks replay output against real-world publication surfaces. For Finland, that means comparing replayed point-in-time text against the Finlex updated/derived comparison surface.

The core rule: Benchmark scores are a proxy. Manual residual review against primary sources is the real verification loop. Similarity scores become interpretable only after the divergence type is classified.

Some divergences enter candidate review when primary-source checks support further investigation beyond the comparison surface. The residual taxonomy classifies each mismatch so that evaluation is not a single number but a typed evidence surface.

August 2026 report

MetricValue
ScopeProject’s Finnish law replay benchmark
ComparisonDeclared Finlex updated/derived surface
Mean normalized text edit distance0.23%
ProvenanceProject-reported development result

Reported in August 2026. The 0.23% measures distance to the declared comparison surface. Correctness and proof coverage require source review, residual classification, and separate denominators.

Evidence package status

The Finnish evidence work exposes replay-vs-Finlex comparison-surface divergences for triage. The project reported 22 candidates that passed its internal review threshold for external review. These remain candidates pending confirmation by a competent authority.

The internal review dataset records statutes, affected provisions, source evidence, comparison role, and disposition. Individual review packets are not part of this website; the public surface reports aggregate status and review boundaries only.

Run the current benchmark

uv sync
uv run lawvm bench --mode official_consolidation --corpus data/finland/bench_core.csv --no-save

This runs the included Finnish benchmark manifest without writing benchmark history or run artifacts. A newly acquired archive can differ as publisher-controlled comparison surfaces change.

The source archive (finlex.farchive) is built from Finlex open data batch downloads. The acquisition scripts and benchmark tooling are in the repository.