GitHub Systems-Code Intelligence Index — Permissive License · 339 Repos · 1,020 PR Harvests
Overview
GitHub Systems-Code Intelligence Index — Permissive License · 339 Repos · 1,020 PR Harvests is a text dataset. Typical applications include code generation. Relevant sectors: Education and AI & Machine Learning. The listing mentions timestamps. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Listing mentions timestamps.
EX-NIHILO-CODE-001 is a commercially gated, permissive-license index of public GitHub repositories and their merged pull request diffs, frozen on 2026-09-16. Built for AI labs, SWE benchmark teams, and code intelligence researchers who need qualified, license-clean systems code — without cloning, without copyleft exposure, and without building the pipeline themselves. WHAT'S INCLUDED Three-layer structured dataset: Layer 1 — Repository Index (3 CSV files) - Full: 339 repositories passing all qualification gates - Verified: 319 repositories with ≥ 100 confirmed merged GitHub PRs - Premium: 205 repositories with ≥ 1,000 stars and ≥ 250,000 estimated LOC Layer 2 — Targeting CSV - Default branch, CI flags, test directory detection - Last 50 merged PRs per repo (title, body, mergedAt, additions, deletions, changedFiles, baseRefOid, headRefOid) Layer 3 — PR Harvest (1,020 PR pairs) - files.json: full GitHub file object arrays per PR (filename, status, patch) - metadata.json: repo name, PR number, language, harvest timestamp - 205 premium repos × 5 PRs each = 1,020 total pairs QUALIFICATION METHODOLOGY Every repository passes five hard gates: - License: MIT, Apache-2.0, BSD-2-Clause, or BSD-3-Clause only — copyleft (GPL, AGPL, LGPL, MPL) is excluded - LOC: 100,000 – 10,000,000 estimated lines (linguist bytes ÷ language-specific BPL divisor) - Language share: primary language ≥ 40% of linguist bytes — polyglot dumps excluded - Recency: pushed within 2 years of index date - Identity:
Technical specifications
Related datasets
More in Code & Software →Sundanese–Indonesian Code-Switching Conversational Speech Corpus — First-of-its-Kind, Consented
The language of West Java is not a language — it is two, braided sentence by sentence.
- Language
- Indonesian
- Volume
- —
- Domain
- Education +1
- Use case
- Speech recognition
More text & language datasets
All 776 text & language datasets →- Global Design Rights Data - the provider Design Registers
- Global Tailored Web Data | AI Training Data | Machine Learning (ML) Data | Tailored Web Data
- Global Trademark Data | 202M+ Trademarks from 198 Authorities
- Government Service Helpline Text Dataset for Energy Analysis (Singapore, English)
- Granular Symptom, Condition, and Risk Factor Annotation
- Greek Literary Book Dataset
- Greek public-domain books, structured long-form, multi-domain
- Greek Speech-to-Text Transcribed Audio Dataset
- Greek Spontaneous Speech Data | 600 Hours | Audio Data
- Gulf Arabic Doctoral Exam Questions and Solutions