Open foundational assets for studying human and AI judgment
AIO Commons publishes the vocabulary for writing down a judgment, the methods for measuring it, and the reference data to compare it against. All three are released in full under CC BY 4.0, so anyone can cite and reproduce them.
Expression, measurement, comparison
The vocabulary of judgment
The common vocabulary for writing a judgment as one line. It covers value, evidence, and source codes, three context axes with 96 cells, and 22 domains.
How judgment is measured
What is measured with that vocabulary, and how. It includes the formula registry and the raw responses of an actual model measurement.
What it is compared against
The point of comparison for a measurement. It records the normative priority relations observed in statutes and precedents.
The downloadable dataset is the September 2026 statute grounding (kr-lnpd@0.9). The current Korea hierarchy (the unified hierarchy: 523 statute articles, 5,857 court decisions) is shown on the observation map.
Published datasets
| Dataset id | Title | Kind | Version | Status | License | Records | Review |
|---|---|---|---|---|---|---|---|
| aio-vocabulary | AIO Judgment Vocabulary | Language | 1.0 | stable | CC-BY-4.0 | 19values | — |
| aio-formulas | AIO Formula Registry | Methods | 1.1 | stable | CC-BY-4.0 | 22formulas | — |
| kr-lnpd | Korea Legal Normative Priority Dataset | Reference | 0.9 | preview | CC-BY-4.0 | 579rules | not started · operator-approved |
| measurement-solar-pro-4-2026-09-02 | AI Measurement — Solar Pro 4 (2026-09-02) | Measurement | 1.0 | experimental | CC-BY-4.0 | 142,918responses | — |
| aio-scenarios | AIO Scenario Library | Scenario | 0.1 | preview | CC-BY-4.0 | 142,918items | — |
| aio-valuemap | Model Response Observation Map Data | Derived | 0.1 | preview | CC-BY-4.0 | 571,672responses | — |
How to read this
Rule and general-relation figures for the reference dataset exclude the base layer (BASE). BASE holds 43 rules and 41 general priority relations: the base hierarchy a national legal system states in common across every domain. It is counted separately and excluded from measurement comparison and headline counts.
The Review column counts the expert review each rule has received. A rule with no review record is operator-approved, which does not mean anyone outside has reviewed it. Disagreements are counted separately rather than hidden, though they stay out of the headline figures. Join the review
Catalog generated at 2026-09-22T20:09:51+09:00 · index.json
Legal professionals take part
Once their credentials are checked, judges, prosecutors, attorneys and legal scholars record their judgments directly on the document cards of statutes and court decisions. There is no operator approval step, and how each credential was checked is published alongside.
Expert interpretation — not a holding of the courtJudgments on the situation axes are the legal professionals’ interpretation. More legal professionals agreeing does not make a priority relation stronger.
How legal professionals take part Working guide Join the review
What is open and what is not
The vocabulary, the methods, and the reference data are released in full under CC BY 4.0. For scenarios, only the calibration set is published; the measurement set is kept closed and rotated to prevent contamination of training data. The only exceptions are three measurement-set items approved as observed cases, released with their original text and stored response (see the release rule in the scenario library).
Commercial protection sits with the organizational measurement service, not with the data. Raw measurement responses carry only record IDs and choices, so no scenario text is released with them.
Citation form
To cite a single record, give the dataset id, the version, and the record id. Record IDs reuse the source data's join keys unchanged: scenario custom_id, stress ledger_id, rule rule_id.
{dataset_id}@{version} / {record_id}
e.g. kr-lnpd@0.9 / KR-MED-R-007 · measurement-solar-pro-4-2026-09-02@1.0 / L4_MED_3-3_2_Sep_Unn_k1_pv1