AI Advisory Notice

migra-check uses large language models, including Anthropic's Claude, to analyze SQL migration scripts and produce risk reports. This page explains — in plain language — what those reports are and are not.

Advisory Only

Risk reports produced by migra-check are advisory. They are not a guarantee of safety, correctness, performance, or any particular outcome. A "low risk" rating does not mean a migration is safe to run. A "high risk" rating does not mean a migration will fail. Ratings reflect a model's best estimate based on the information you provide and the knowledge embedded in the model and prompts.

AI Can Be Wrong

Like all large language models, Claude can produce incomplete, outdated, or confidently incorrect output. It may miss risks that a human DBA would catch. It may also flag risks that do not apply in your specific environment. Analyses depend on the accuracy of the schema metadata you provide and on context that may not be visible to the model (application behavior, traffic patterns, custom tooling, replica topology, and more).

You Remain Responsible

You are solely responsible for:

• Reviewing every finding and recommendation in a risk report.
• Testing migrations in a non-production environment that mirrors production as closely as practical.
• Deciding whether and when to run a migration against a production database.
• Having rollback and recovery procedures in place.
• Complying with change management, approval, and audit requirements that apply inside your organization.

No Warranty, Limited Liability

Use of the Service is governed by our Terms of Service, including the disclaimers of warranty and limitation of liability set out there. Nothing on this page expands our obligations beyond what those Terms provide.

Do Not Submit Secrets or Personal Data

Migration scripts and schema metadata you submit are sent to our AI subprocessor (Anthropic) for processing. Do not include secrets, credentials, production row data, or personal data in submissions. See our Acceptable Use Policy for details.