v1.2.2 · Updated July 2026

Dataset Card

Comprehensive documentation for the SWE-Enterprise dataset — structure, quality, licensing, and citation.

Overview

SWE-Enterprise is a synthetic dataset of 24,316 fully simulated enterprise software companies, purpose-built for training and evaluating AI coding agents, RAG pipelines, and fine-tuning models. Each world is a self-contained directory containing multi-repository codebases, team Slack conversations, injected software bugs, engineering missions, architecture decision records, and narrative company metadata.

24,316

Enterprise Worlds

97,290

RAG Tasks

10

Domains

100%

Fact Accuracy

Data Structure

Each world is organized as a standardized directory with the following components:

Source Code

30–60 files across 2–4 repos in JavaScript, TypeScript, and Python. Organized by microservice boundaries with realistic stub implementations.

Slack History

Multi-participant conversations covering production incidents, feature planning, and architecture debates. Threads reference specific code files.

Injected Bugs + Missions

1–3 bugs per world at critical and medium severity. Three missions per world with the first targeting the injected bug.

Architecture Decision Records

ADRs explaining database selection, service boundaries, caching strategies, and other technical decisions.

Company Metadata

Narrative company description, org structure, engineering culture, active initiatives, and persona profiles.

RAG Evaluation Tasks

Three task types: retrieval (find relevant code from Slack), reasoning (multi-step bug diagnosis), and grounding (factual QA).

Quality & Validation

140 Automated Tests

12 test files covering label correctness, schema validity, org name uniqueness, RAG export integrity, and bug-category balance.

Label Correctness Audit

Two-layer audit: rule-based checks on all 97,290 tasks + automated AI-assisted verification on a stratified sample. 100% structured fact accuracy.

ChatGPT IP Audit

Independent third-party review of all documentation and sample data. Final score: 10/10. IP risk: 1/10.

Split Integrity

Deterministic train/val/test split (70/15/15) via SHA-256 hash. Zero overlap across splits. Fully reproducible.

Human Review

50-world manual spot check confirmed structural correctness, narrative coherence, and mission-bug alignment.

Continuous Monitoring

Competitor market intelligence platform tracks Kaggle, HuggingFace, and AWS Marketplace for comparable datasets.

License

SWE-Enterprise is distributed under a Commercial Training License. This license grants you the right to:

  • ✓ Use the dataset to train machine learning models
  • ✓ Deploy trained models in production environments
  • ✓ Benchmark and evaluate model performance
  • ✓ Publish research results and benchmark scores
  • ✓ Distribute derivative models (not the raw dataset)

The dataset itself may not be redistributed. Community and Research editions use a separate non-commercial license. See LICENSE.txt in each edition package for full terms.

Citation

@misc{ajaxspeaks-swe-enterprise-2026,
  title     = {SWE-Enterprise: 24,316 Synthetic Enterprise Worlds
               for AI Training and Benchmarking},
  author    = {AjaxSpeaks},
  year      = {2026},
  month     = {July},
  note      = {Version 1.2.2},
  howpublished = {\url{https://ajaxspeaks.com}}
}

A CITATION.cff file is included in all edition packages.

Downloads

Free Sample

One complete world with all components. Try before you buy.

Download →

QA Report

Full quality assurance report with test results, audit methodology, and limitations.

Coming soon

Documentation Bundle

Dataset card, license, citation, reproducibility guide, and changelog.

Coming soon

Editions

From free samples to enterprise-wide deployment. All editions include full Commercial Training License.

Feature Evaluation Research Professional Enterprise
5 Preview Worlds ✓ ✓ ✓ ✓
1,000 Worlds — ✓ ✓ ✓
5,000 Worlds — — ✓ ✓
Full 24,316 Worlds — — — ✓
Commercial License — ✓ ✓ ✓
QA Report ✓ ✓ ✓ ✓
Dataset Updates — 6 months 12 months Ongoing
Priority Support — — ✓ ✓
Evaluation
5 worlds · Evaluate quality before you commit.
  • 5 enterprise worlds
  • Sample from all 10 domains
  • All 5 components per world
  • Unrestricted evaluation
Request Quote
Research
1,000 worlds · For individuals, universities, and startups.
  • 1,000 enterprise worlds
  • All 10 domains represented
  • 1,000+ RAG evaluation tasks
  • Full commercial training license
  • Documentation bundle
Request Quote
Professional
5,000 worlds · Built for commercial AI teams.
  • 5,000 enterprise worlds
  • Stratified domain sampling
  • 20,000+ RAG evaluation tasks
  • Full commercial training license
  • Priority support
Request Quote
Enterprise
Full dataset · For organizations requiring maximum coverage.
  • 24,316 worlds (full dataset)
  • Maximized domain + difficulty coverage
  • 97,290 RAG evaluation tasks
  • Full commercial training license
  • Custom domain weighting available
  • Dedicated support + onboarding
Request Quote

We use minimal essential cookies for site functionality. No third-party tracking. Learn more.