The same method as our services, as tools you run: grade a deletion claim, test a removal method, audit a production model with planted canaries, and catch a removal that stops holding after the next training run.
For anyone holding a deletion or unlearning claim
Grade your own evidence against the Standard in your browser. Your counts never leave the page. Self-assessments cap at Tier 3 by design.
Grade your evidence →For auditors, buyers and builders
The open, versioned protocol every Yuyari tool and verdict follows: what a removal claim must survive, and what each verdict licenses you to say. CC BY 4.0.
Read the Standard →These are implemented and in final validation. Join the waitlist below and we'll tell you when each one ships.
For ML teams with a GPU
Rehearsal: install a fictional target, remove it with your method, and attack it. Canary audit: plant fictional canaries in fine-tuning, remove them with your real deletion set, and audit the production model against the checkpoint you started from.
For assessments you'll rely on
Fresh, private fictional packs minted per assessment, so no one can train against the probes. A self-assessment tier, and a verified tier where Yuyari holds the probes back until scoring.
For MLOps and platform teams
A removal that held can stop holding after the next fine-tune, merge or quantisation. A GitHub Action and GitLab template re-audit your canaries on your own GPU runner and fail the build when the removal regresses.
For teams tracking models in MLflow
Run an audit with mlflow.evaluate and log the verdict, its intervals and the evidence
beside the model they describe.
Need the decisive test now? A Rehearsal Engagement runs the full protocol on your process, by application.