Overview

⚠️ Pre-alpha (v0.0.1). Designed for individual researchers who want to version, restore, and share datasets. Sharing via push/pull works today. Simultaneous editing is not supported.

The Problem

You’ve got a 500 GB fMRI dataset. You need to:

• Track which files belong to which analysis
• Save and restore checkpoints as preprocessing changes
• Rename and restructure without copying 500 GB
• Share data with collaborators
• Verify data integrity months later

Git can’t handle 500 GB files. Git-annex can, but it adds significant complexity — it wraps git internals with adjusted branches, direct mode, and special remotes. DVC needs a git repo and pipeline definitions. Rsync gives you copies, not versions.

These tools weren’t designed around the workflow of versioning research datasets directly.

What GDA Does

GDA versions research data. You add files, it hashes them, stores them by fingerprint, and replaces originals with symlinks. That’s it.

gda init                    # Start tracking
gda add bids/               # Add files (instant dedup)
gda snapshot preprocessing  # Save a checkpoint
gda checkout raw            # Restore previous state
gda push origin             # Sync to remote storage

Why It’s Different

No merge conflicts (because there’s no merging)

GDA has no branches. You work on one dataset state at a time. When you checkout, GDA replaces your entire project folder — there’s nothing to merge.

This is intentional for researchers working alone: you don’t want to resolve a conflict in a 10 GB NIfTI file. Snapshot before a risky step, checkout if it goes wrong.

If you need parallel dataset versions (e.g. different preprocessing streams), that’s branches — planned but not yet implemented.

No git required

GDA identifies files by a fingerprint (SHA256) computed from their content, not by their filename or path. Add the same file twice — stored once. Move or rename — instant, because only the symlink changes.

Snapshots, not commits

A snapshot records every file’s fingerprint at a point in time. It’s a bookmark, not a diff. No commit graph, no history rewriting, no merge commits.

Integrity by design

Stored files are read-only and named by their fingerprint. gda fsck verifies every file against its hash. If something corrupted, you’ll know immediately.

Is GDA Right For You?

Good fit:

• You download datasets and want to track preprocessing steps
• You share data with labmates and need everyone on the same file tree
• You want to archive data with integrity verification
• You restructure your files constantly and hate waiting for copies
• Git-annex complexity is a barrier for your workflow

Not a good fit (yet):

• You need two people editing the same dataset at the same time (branches/merging planned)
• You need S3 or HTTP remotes (rsync only for now)
• You need Windows support (planned)

Comparison

	GDA	git-annex	DVC	Git LFS
No git needed	✓	✗	✗	✗
Conflict-free (no branches)	✓	✗	✗	✗
Concurrent editing	✗	∼ limited	✓	✓
Deduplicates identical files	✓	✓	✓	✓
Symlink file listing	✓	✓	✗	✓
Remote sync method	rsync	S3, rsync, many	S3, GCS	S3, GH

Why no concurrent editing? GDA has no branches and no merge mechanism. It’s designed for one working tree at a time — perfect for a single researcher or for sharing data where only one person modifies at a time. If you need two people editing the same dataset simultaneously, this isn’t ready for that yet. Branches and merging are planned.

→ Installation — get GDA on your machine → Quick Start — track your first dataset in 5 minutes