Finland
The empirical proving ground.
Why Finland?
Finland combines difficult replay dynamics with unusually open source material and a legally non-authoritative consolidation surface. Finlex consolidated texts are informational, not legally binding. The "real" law is the original statute plus all amendment acts published in Säädöskokoelma.
This makes replay-vs-oracle comparison scientifically meaningful: when LawVM and Finlex disagree, the disagreement can be independently adjudicated against primary sources. A divergence might be a LawVM bug, a source gap, or a genuine error in the official consolidation.
Corpus
690 Finnish statutes with at least one amendment, spanning from the 1920s to the 2020s. All replayed from raw amendment acts in Säädöskokoelma XML. Curated from 3,591 amended statutes — the 690 are those where base XML, oracle consolidation, and all amendment texts are available for full replay.
| Decade | Statutes | Avg amendments | Max |
|---|---|---|---|
| 1920s–1960s | 10 | 15 | 67 |
| 1970s | 20 | 13.7 | 61 |
| 1980s | 25 | 14.4 | 64 |
| 1990s | 107 | 20.6 | 238 |
| 2000s | 173 | 13.3 | 166 |
| 2010s | 237 | 5.3 | 52 |
| 2020s | 118 | 2.5 | 18 |
As of 2026-04-16.
Benchmark
Text distance
Structural error
<90% match
Benchmark run: 2026-04-16, mode: finlex_oracle. Two metrics: Levenshtein measures character-level text distance; structural error measures section-level divergence. See Artifacts for corpus definition and reproducibility.
The golden dataset
We systematically investigate every divergence between LawVM and Finlex. In 77+ cases so far, the official Finlex consolidation contains errors — missing content, stale amendments, editorial additions without legal basis, missing section headings.
Each finding is documented in a structured YAML entry with Finnish prose, root cause classification, affected sections, and source evidence. The full findings database is available in Finnish at Finlex-virheet.
Root cause taxonomy
| Root cause | Count | Meaning |
|---|---|---|
| finlex_missing_content | 10 | Finlex omits content published in Säädöskokoelma |
| source_pathology | 9 | Source XML itself is defective |
| finlex_missing_otsikko | ≥9 | Finlex omits section headings |
| oracle_stale_cutoff | ≥6 | Finlex has not applied recent amendments |
| finlex_editorial_addition | 4 | Finlex added content without amendment basis |
| high_uncovered_body | 3 | Structural amendment coverage gap |
| finlex_dual_element_residual | 3 | Finlex has dual current/pending versions |
| corrigendum_misapplied | 1 | Published correction not applied |
| + 7 more categories |
77 entries as of 2026-04-16. Target: full classification of all divergences.
Concrete cases
2004/699 — Wrong section heading for 16+ years
Laki rahoitus- ja vakuutusryhmän valvonnasta. Amendment 2008/886 changed §8's heading from “Valvontatehtävän siirtäminen toiselle valvontaviranomaiselle” to “... ulkomaan valvontaviranomaiselle” (“another” → “foreign” supervisory authority). Finlex still shows the 2004 heading. No subsequent amendment changed it back. A semantic error that changes legal meaning, persisting in the official database since 2008.
2014/716 — Missing COVID-19 emergency provisions
Valtioneuvoston asetus yritystoiminnan kehittämiseksi. Temporary COVID-19 amendment 2020/697 modified §§5, 11, 12 and added new §8b (state aid for enterprises in difficulty). Finlex's consolidated version contains none of these changes — it shows pre-COVID text even though the consolidation was prepared during the amendment's period of force. Anyone relying solely on Finlex during 2020–2021 was reading law that omitted applicable emergency provisions.
2014/1245 — Six section headings silently dropped
Valtioneuvoston asetus televisio- ja radiotoiminnasta. Amendment 2018/396 added headings to §§1, 3, 6, 7, 14, 15. Finlex's consolidated version is missing all of them. The amendment source XML explicitly contains the headings. Not a subtle text difference — six headings simply absent from the official version.
1992/728 — Future-effective amendment already applied
Laki kunnallisesta viranhaltijasta. Finlex's consolidated metadata says the consolidation date is 2009-12-29, but §3 already reflects amendment 2009/1710 which enters into force on 2010-01-01. LawVM's legal point-in-time mode correctly carries the earlier wording. The editorial consolidation runs ahead of strict legal effectivity — silently incorporating provisions that are not yet in force.
1997/1339 — Corrigendum applied editorially but not in machine-readable source
Kirjanpitoasetus. Amendment 2015/1752 had a published corrigendum correcting multiple preamble errors. Finlex applied the correction; the machine-readable enacted XML was never updated. LawVM's strict-from-source replay is transparent about what it does and does not have. A mixed case illustrating that corrigenda exist outside the machine-readable pipeline.
The forward goal
Current evidence already shows repeated classes of cases where replay outperforms the official consolidation. The ongoing goal is a fully classified divergence ledger for the Finnish statute corpus — every divergence investigated, every one typed.