meta-introspector: Unified Code Analysis Dataset

Organization: introspector
Dataset: meta-introspector
License: AGPL-3.0

🎯 Dataset Summary

The meta-introspector dataset provides a comprehensive index of 3.66 million files with full git provenance tracking. This dataset enables:

• Code search across millions of files
• Provenance tracking from files to git commits
• Repository analysis at scale
• ML training on real-world code data

📊 Dataset Structure

Data Instances

Each row in indexes/files.parquet represents a file with its git metadata:

{
  "file_path": "/mnt/data1/meta-introspector/src/main.rs",
  "git_repo": "/mnt/data1/meta-introspector",
  "commit": "7fc28d09...",
  "branch": "novel-code-analysis-clean",
  "remote": "git@github.com:meta-introspector/meta-introspector.git",
  "url": "https://github.com/meta-introspector/meta-introspector/blob/novel-code-analysis-clean/src/main.rs",
  "tracked": true
}

Data Fields

• file_path (string): Absolute file path
• git_repo (string): Git repository root path
• commit (string): Git commit hash (40 chars)
• branch (string): Git branch name
• remote (string): Git remote URL
• url (string): GitHub/GitLab URL for the file
• tracked (boolean): Whether file is tracked in git

Data Splits

Single dataset with 3,660,152 files (no train/test split).

📈 Dataset Statistics

• Total files: 3,660,152
• Files with git repo: 2,366,334 (64.7%)
• Tracked in git: 1 (0.0%)
• Untracked: 2,366,333 (64.7%)
• File size: 437MB (Parquet, compressed)
• Compression ratio: ~6:1 (vs 763MB CSV)

🚀 Usage

Load with Pandas

import pandas as pd

# Load dataset
df = pd.read_parquet('hf://datasets/introspector/meta-introspector/indexes/files.parquet')

# Find Rust files
rust_files = df[df['file_path'].str.endswith('.rs')]
print(f"Found {len(rust_files)} Rust files")

# Find files by repository
repo_files = df[df['git_repo'].str.contains('meta-introspector')]
print(f"Found {len(repo_files)} files in meta-introspector")

Query with DuckDB

import duckdb

# SQL queries on Parquet
con = duckdb.connect()
result = con.execute("""
    SELECT git_repo, COUNT(*) as file_count 
    FROM 'hf://datasets/introspector/meta-introspector/indexes/files.parquet'
    WHERE git_repo != ''
    GROUP BY git_repo 
    ORDER BY file_count DESC 
    LIMIT 10
""").fetchdf()
print(result)

Load with Polars

import polars as pl

# Fast columnar processing
df = pl.read_parquet('hf://datasets/introspector/meta-introspector/indexes/files.parquet')

# Filter and aggregate
rust_by_repo = (
    df.filter(pl.col('file_path').str.ends_with('.rs'))
      .groupby('git_repo')
      .agg(pl.count().alias('rust_files'))
      .sort('rust_files', descending=True)
)
print(rust_by_repo.head(10))

🎓 Use Cases

Code Search

Find files across 3.66M indexed files with full git context.

Provenance Tracking

Trace any file to its git commit, branch, and remote repository.

Repository Analysis

Analyze code distribution, file types, and repository structures at scale.

ML Training

Train models on real-world code data with git metadata for context.

Research

Study large-scale code repositories and their evolution.

🔬 Dataset Creation

Source Data

Files were collected from /mnt/data1 filesystem using parallel scanning with 24 workers.

Data Collection

1. Filesystem scan: 3.66M files discovered
2. Git mapping: Each file associated with git repository
3. Metadata extraction: Commit hash, branch, remote URL
4. Tracking status: Checked if file is tracked in git
5. Parquet export: Columnar format with compression

Performance

• Scan time: 12 seconds
• Throughput: ~305,000 files/second
• Workers: 24 parallel (rayon)
• Caching: Git info cached per repository (not per file)

Tools Used

• git_file_mapper.rs - Rust tool for parallel file scanning
• rayon - Parallel processing
• arrow + parquet - Columnar storage

📊 Dataset Characteristics

File Types

The dataset includes all file types found in the scanned repositories:

• Source code (.rs, .py, .js, .c, .cpp, etc.)
• Configuration files (.toml, .json, .yaml, etc.)
• Documentation (.md, .txt, .rst, etc.)
• Build artifacts
• Binary files

Repository Coverage

Files from 1,000+ git repositories including:

• Open source projects
• Local development repositories
• Nix store derivations

🔗 Related Datasets

• solfunmeme-index - 1.2M+ Rust semantic analysis records
• git-activity - Git activity tracking

🤝 Contributing

This dataset is automatically updated from the meta-introspector project.

To contribute:

1. Run git_file_mapper from the project
2. Generate data/indexes/files.parquet
3. Submit updates via pull request

📄 Citation

@dataset{meta_introspector_2026,
  title={Meta-Introspector: Unified Code Analysis Dataset},
  author={Meta-Introspector Team},
  year={2026},
  url={https://huggingface.co/datasets/introspector/meta-introspector},
  note={3.66M+ files with git provenance and comprehensive analysis}
}

📝 License

AGPL-3.0 - See LICENSE file for details.

---

Generated: 2026-01-17
Project: https://github.com/meta-introspector/meta-introspector
Organization: https://huggingface.co/introspector