Trust Center
Review how Tuned Predictions handles model evaluation, versioning, and human review, plus the data and security questions to discuss before getting started.
Security and privacy
SOC audited. GDPR compliant.
Contact our team for audit documentation and data protection questions.
Review the data before training
The workflow starts with connected data and prepared examples. Review the inputs and target outcomes before training. Use only information available at prediction time. See the product workflow and supported integrations.
Evaluate before deployment
Use held-out examples, per-question results, and available baseline comparisons to judge a candidate. Training does not guarantee an improvement. A scheduled retraining run creates a candidate; evaluation and deployment remain separate steps.
Follow a prediction back to its version
Versioned inference and traces help you inspect the inputs and model version behind a result. Keep the task and evaluation criteria clear when comparing versions.
Questions for a security review
Contact the team about access controls, retention and deletion, hosting locations, model provider data use, contractual terms, or audit documentation. Describe your planned use and the information you need reviewed.
You decide how to use a prediction
Predictions are estimates, not guaranteed outcomes. Review them alongside the available evidence and your own judgment. The examples on this website use fictional inputs and illustrative outputs.