ToetssteenRegister van rechterlijke toetsing
English summary

Method, sources and limits

Toetssteen is a public register of Dutch judicial review of asylum decisions. This page summarises how the data is collected, how the story types are assigned, and what the figures cannot tell you.
Primary source
rechtspraak.nl open data
Unit of analysis
Published ruling (ECLI)
Classification
Automated, with source passage
Language of sources
Dutch
01Pipeline

From published ruling to figure

  1. 01

    Collection

    All published rulings on asylum decisions are retrieved from the Dutch judiciary's open data API (rechtspraak.nl), including case number (ECLI), court, date, summary and full text where available.

  2. 02

    Structured extraction

    Each ruling is parsed into fixed fields: procedure, country of origin, outcome for the immigration service, reasoning style, and whether country-of-origin information was weighed. Every extracted value is stored with the passage it came from.

  3. 03

    Story-type classification

    Where a ruling assesses the asylum account itself, a language model assigns one archetype (conversion, LGBTI orientation, political activity, family ties, concealed origin, document fraud, medical grounds, draft evasion, military service, blood feud) plus the credibility indicators the court relied on. Each label carries a quoted source passage.

  4. 04

    Aggregation

    Counts and upheld rates are computed per archetype, per country of origin and per year. The denominator is always the number of judged cases, excluding partial and unclear outcomes.

  5. 05

    Per-case probability

    A case-level estimate combines the base rate with archetype, country and procedural features using Laplace smoothing, so small cells do not produce extreme values. It is a descriptive statistic about published case law, never a prediction about an individual person.

BronCourt rulings are published by the Dutch judiciary under an open licence. Personal data is already redacted at source; Toetssteen adds no identifying information.

02Limits

What the register does not measure

Selection at two stages. Only decisions that are contested reach a court, and only part of those rulings is published. Granted asylum applications never appear in the data.

Outcome is legal, not factual. A quashed decision means the court found the reasoning insufficient or the decision unlawful; it does not establish that the account was true, nor the reverse.

Automated labels carry error. Classification is machine-assigned with a quoted passage and spot-checked editorially. Confidence intervals per archetype are published on the Dutch pages.

No individual prediction. The per-case estimate describes patterns in published case law. It cannot and must not be used to judge a person.

Intended use
The register is built for journalists, researchers, lawyers and policymakers who want a traceable figure instead of an impression. Every number links back to an ECLI.
Source language
Court rulings, quoted passages and case summaries are in Dutch, as published by the Dutch judiciary. These English pages summarise the figures and the method. The full detail lives on the full Dutch methodology (in Dutch).
Probability model in detail (in Dutch)Data quality audit (in Dutch)