Gather the threads
Read folders, JSON, web sources, and generated records.
The loom is open
Weave raw data
into a training release.
Bring every source, rule, split, and training target together in one repeatable build.
More than storage
Define the exact weave that becomes training.
Your source data can stay where it lives. BRAID selects the exact records, joins them under one contract, and produces the release your model actually sees.
Prepare the weave
Fix the format, split long text, and keep the source link.
Apply the rules
Check rights, credit, safety, and quality before use.
Remove copies
Catch exact copies and close matches before training.
Choose the best
Build a repeatable mix and balance the topics you need.
Prove it works
Test the size, quality, topics, splits, and budget.
Ship it
Get ready files, a data card, an audit trail, and checksums.
The finished weave
Ship the corpus and its proof together.
Move the release anywhere. BRAID can check that every file, split, and record is still exactly where it belongs.
release/
├── release.json
├── braid.yaml
├── README.md
├── checksums.sha256
├── data/
│ ├── train.jsonl
│ └── train.parquet
├── audit/
│ ├── membership.jsonl
│ └── decisions.jsonl
└── evaluation.json