GitHub View workflow

Browser assurance verification

Test evidence produced by a model succession.

Choose a recorded candidate, run the evaluation and readiness layers, review the decision, and introduce a controlled failure.

This is not the model-migration engine. The local CLI performs model loading, recovery training, candidate selection, and adapter export. This Lab evaluates frozen or locally uploaded predictions; no model training or fresh inference happens here.

Step 1

Choose a candidate

Results remain hidden until evaluation.

About the selected scenario

This candidate was selected before final evaluation and can be challenged with a controlled failure after the first run.

Open full succession evidence
Advanced toolsUpload predictions, download samples, and export receipts

Local predictions

Evaluate JSON or JSONL

Files stay local. Maximum 5 MiB. Uploads are evaluated separately from frozen evidence and never claim frozen parity or integrity. A complete compatible final set of 96 records can receive readiness; partial compatible sets receive evaluation only.

Run an evaluation or select a file to verify and load the sample asset.