---
language:
- uk
license: cc-by-4.0
task_categories:
- text-classification
task_ids:
- multi-class-classification
pretty_name: "Ukrainian Court Decisions: Case Outcome Prediction (6.7M)"
size_categories:
- 1M<n<10M
tags:
- legal
- court-decisions
- ukraine
- judgment-prediction
- legal-nlp
- temporal-split
- edrsr
- arxiv:2605.14890
- arxiv:2605.15362
- arxiv:2605.17639
source_datasets: []
configs:
- config_name: pre_war
  data_files:
  - split: train
    path: pre_war/train-*.parquet
  - split: validation
    path: pre_war/validation-*.parquet
  - split: test
    path: pre_war/test-*.parquet
- config_name: hybrid_war
  data_files:
  - split: train
    path: hybrid_war/train-*.parquet
  - split: validation
    path: hybrid_war/validation-*.parquet
  - split: test
    path: hybrid_war/test-*.parquet
- config_name: full_scale
  data_files:
  - split: train
    path: full_scale/train-*.parquet
  - split: validation
    path: full_scale/validation-*.parquet
  - split: test
    path: full_scale/test-*.parquet
dataset_info:
  features:
  - name: text
    dtype: string
  - name: label
    dtype: string
  - name: language
    dtype: string
  - name: jurisdiction
    dtype: string
  - name: adjudication_date
    dtype: string
  - name: doc_id
    dtype: string
---

# Ukrainian Court Decisions: Case Outcome Prediction (6.7M)

The largest publicly available dataset of Ukrainian court decisions for **case outcome prediction**, extracted from the [State Court Decisions Registry](https://reyestr.court.gov.ua/) (EDRSR). Contains **6,690,284** substantive decisions from civil and commercial courts spanning 2008--2026, with temporal splits across three wartime epochs.

## Overview

Ukraine's EDRSR is one of the world's largest open judicial databases, containing 100M+ judicial documents. This dataset extracts the 12.6M substantive decisions (judgment_code=3, "Рішення") from civil and commercial jurisdictions, filtering to 6.7M with parseable case outcomes.

For the benchmark-sized version (428K balanced, with LEXTREME configs), see [overthelex/ukrainian-court-decisions](https://huggingface.co/datasets/overthelex/ukrainian-court-decisions).

## Dataset Structure

Three temporal epoch configs with chronological train/validation/test splits (80/10/10):

| Config | Train | Validation | Test | Total | Period |
|--------|------:|----------:|-----:|------:|--------|
| `pre_war` | 434,609 | 54,326 | 54,327 | 543,262 | 2008--2013 |
| `hybrid_war` | 3,349,700 | 418,712 | 418,714 | 4,187,126 | 2014--2021 |
| `full_scale` | 1,567,916 | 195,989 | 195,991 | 1,959,896 | 2022--2026 |

### Fields

| Field | Description |
|-------|-------------|
| `text` | Facts section of the decision (full, untruncated) |
| `label` | Case outcome: `approved`, `dismissed`, or `partial` |
| `language` | Always `uk` (Ukrainian) |
| `jurisdiction` | `civil` or `commercial` |
| `adjudication_date` | Date of the decision (YYYY-MM-DD) |
| `doc_id` | Unique document ID in EDRSR |

### Label distribution (natural, unbalanced)

| Epoch | approved | partial | dismissed |
|-------|---------|---------|-----------|
| pre_war | 78.8% | 16.0% | 5.2% |
| hybrid_war | 76.5% | 14.5% | 9.0% |
| full_scale | 79.0% | 13.2% | 7.8% |

## Temporal Epochs

The three epochs correspond to distinct phases of Ukrainian legal history, with substantial shifts in judicial practice:

| Epoch | Period | Context |
|-------|--------|---------|
| **pre_war** | 2008--2013 | Stable legal environment, ~832 courts operational |
| **hybrid_war** | 2014--2021 | Crimea annexation, Donbas conflict, ~40 courts lost, displaced judiciary |
| **full_scale** | 2022--2026 | Full-scale Russian invasion, nationwide martial law, new legal regimes |

Splits are **temporal** (chronological by adjudication date), not random. Earlier decisions are in train, later ones in test -- simulating real deployment conditions.

## Task

Given the **facts section** (ВСТАНОВИВ) of a court decision, predict the case outcome:

| Label | Ukrainian | Description |
|-------|-----------|-------------|
| `approved` | Задоволено | Claim fully satisfied |
| `dismissed` | Відмовлено | Claim dismissed |
| `partial` | Частково задоволено | Claim partially satisfied |

Only the facts section is provided as input. The court's reasoning and the dispositive (ruling) section are excluded to prevent trivial classification.

## Usage

```python
from datasets import load_dataset

# Load a specific epoch
ds = load_dataset("overthelex/ua-case-outcome-6m", "hybrid_war")

# Cross-epoch evaluation
train = load_dataset("overthelex/ua-case-outcome-6m", "pre_war", split="train")
test = load_dataset("overthelex/ua-case-outcome-6m", "full_scale", split="test")
```

## Related Datasets

- [overthelex/ukrainian-court-decisions](https://huggingface.co/datasets/overthelex/ukrainian-court-decisions) -- benchmark-sized version (428K balanced) with LEXTREME configs
- [overthelex/ua-court-citation-graph](https://huggingface.co/datasets/overthelex/ua-court-citation-graph) -- 2.3M co-citation edges from 99.5M court decisions
- [overthelex/ua-statute-retrieval](https://huggingface.co/datasets/overthelex/ua-statute-retrieval) -- statute retrieval benchmark from 396M citations
- [overthelex/ua-temporal-drift](https://huggingface.co/datasets/overthelex/ua-temporal-drift) -- temporal concept drift analysis data
- [joelniklaus/lextreme](https://huggingface.co/datasets/joelniklaus/lextreme) -- multilingual legal NLU benchmark (includes Ukrainian subset)

## Related Papers

- **Temporal Decay of Co-Citation Predictability: A 20-Year Statute Retrieval Benchmark from 396M Ukrainian Court Citations** ([arXiv:2605.17639](https://arxiv.org/abs/2605.17639)) -- uses the same EDRSR source registry; analyzes temporal effects on legal information retrieval
- **A Citation Graph from 100 Million Court Decisions: Structure, Dynamics, and Applications** ([arXiv:2605.15362](https://arxiv.org/abs/2605.15362)) -- constructs a citation network from the same 100M+ EDRSR documents
- **Tokenizer Fertility and Zero-Shot Performance on Ukrainian Legal Text** ([arXiv:2605.14890](https://arxiv.org/abs/2605.14890)) -- evaluates tokenizer quality on Ukrainian court decisions
- **LEXTREME: A Multi-Lingual and Multi-Task Benchmark for the Legal Domain** ([arXiv:2301.13126](https://arxiv.org/abs/2301.13126)) -- the benchmark this dataset contributes to

## Source Data

- **Registry**: State Court Decisions Registry of Ukraine (ЄДРСР), [reyestr.court.gov.ua](https://reyestr.court.gov.ua/)
- **Document type**: Substantive decisions only (judgment_code=3, "Рішення")
- **Jurisdictions**: Civil (justice_kind=1) and commercial (justice_kind=3)
- **Text processing**: anonymization tokens (ОСОБА_1, АДРЕСА_1, etc.) normalized to [PERSON], [ADDRESS], [NUMBER], [INFO]
- **Outcome extraction**: regex-based classification from the dispositive section

## Citation

```bibtex
@misc{ovcharov2025uacaseoutcome,
  title={Ukrainian Court Decisions: Case Outcome Prediction Dataset (6.7M)},
  author={Ovcharov, Volodymyr},
  year={2025},
  url={https://huggingface.co/datasets/overthelex/ua-case-outcome-6m},
  note={Extracted from the State Court Decisions Registry of Ukraine (EDRSR)}
}
```

## License

CC-BY-4.0. The source data is published by the State Court Administration of Ukraine under open access provisions per Ukrainian law "On Access to Court Decisions" (2005).