#Exporting & Sharing Results

eval_752 exports evaluation results as .eval752.zip bundles — complete, self-contained packages that capture everything about a run.

#What's in a bundle?

An exported .eval752.zip contains:

• Dataset snapshot — the exact items that were evaluated
• Run configuration — provider, model, parameters
• Per-item results — each response, score, latency, and error
• Metadata — timestamps, eval_752 version, environment info
• Assets — any embedded images or files

This is the format for reproducibility. When you share a bundle, the recipient can inspect every detail without needing access to your instance.

#How to export

1. Go to Runs
2. Open a completed run (from the active board or archive)
3. Open the inspector
4. Click Export
5. Download the .eval752.zip file

#When to export

• Sharing evidence — "Here's proof that Provider X scored 15% lower than Provider Y on this benchmark"
• Archiving for compliance — keep a frozen record of evaluation results
• Handing off — giving results to someone who runs their own eval_752 instance
• Auditing — the bundle contains raw responses, not just aggregate scores
• Debugging — unzip locally and inspect results.jsonl for item-level data

#Re-importing

An exported bundle can be imported into another eval_752 instance using the dataset upload flow (Datasets → Upload eval752 package). The import recreates the dataset and, if results are included, the completed run.

#What's not available yet

• Direct CSV export from the browser
• Direct JSON export from the browser
• Push to Hugging Face Hub from the UI
• PDF or HTML report generation

For now, if you need raw data, export the .eval752.zip and inspect the contained JSONL and JSON files directly.

See Reproducibility for more on what bundles capture and how to think about evidence quality.