---
license: apache-2.0
task_categories:
- text-generation
language:
- en
tags:
- gherkin
- bdd
- synthetic
pretty_name: 'Gherkin EN 100k '
size_categories:
- 100K<n<1M
---

# Gherkin EN 100k

**Summary:**  
English-only, 100k-sample synthetic dataset for converting free-form requirements/paragraphs to **Gherkin** (BDD) feature files. Generated in a **GPT‑5 style** (no chain-of-thought), with business-facing, testable scenarios.

## Contents

- 5 shards × 20k each (JSONL), anonymized (no real personal names; neutral phrase **"the user"** used instead)
- Each line schema:
  - `input_text` (string): free-form requirement/spec paragraph
  - `gherkin` (string): valid Gherkin with `Feature`, optional `Rule`/`Background`, 2–4 `Scenario`s or a `Scenario Outline` with `Examples`
  - `task` (string): `"gherkin_conversion"`
  - `language` (string): `"en"`
  - `tags` (list[str]): e.g., `@smoke`, `@regression`, `@api`, `@compliance`
  - `difficulty` (enum): `easy|medium|hard`
  - `domain` (enum): `ecommerce|banking|healthcare|travel|food_delivery|education|media|hr|analytics|saas`
  - `created_utc` (RFC3339 string)
  - `source` (string): `"synthetic-gpt5-style"`

## Statistics
- **Total examples:** 100000
- **Shards:** 5 × 20k
- **Approx size:** 137.3 MB (JSONL raw)

## Files

- gherkin_en_20k_part1.jsonl
- gherkin_en_20k_part2.jsonl
- gherkin_en_20k_part3.jsonl
- gherkin_en_20k_part4.jsonl
- gherkin_en_20k_part5.jsonl
- gherkin_en_100k_manifest.json


## Intended use

- Train/finetune **student models** via **distillation** for **requirements → Gherkin** conversion.
- Pretraining/adapter training for BDD-aware assistants.

## License
```
apache-2.0

```
## Citation

If you use this dataset, please cite:
```
Kavindu Hansaka Jayasinghe (2025). Gherkin EN 100k.
```