AlphaIQ · CertifyIQ
Model validation evidence a supervisor can check.
CertifyIQ checks models — statistical, machine-learning or GenAI — and writes down the proof. Every number comes from evidence you can trace, every run is pinned to the exact data and tool version used, and every certificate says clearly what wasn't tested, not just what was.
Try the demo →
See a real validation run on sample data for a credit-default model. No account needed, nothing to install.
See an engagement →
Look at the same evidence a validator sees: the list of issues found, what's needed before sign-off, and the final report.
What it checks →
What gets measured for each kind of model, and — just as clearly — what this tool doesn't check yet.
Staff sign-in →
AlphaIQ's internal pages are only for approved staff. Sign in if that's you.
Why you can trust the evidence
- You can reproduce every result. Run it again on the same data and you get the same numbers back, every time.
- We say when we can't check something. If a number can't be computed, the report explains why instead of just leaving it out.
- Certificates say what was tested, not whether the model is "good". They list what was checked and what wasn't, and stop being valid the moment the model changes.
- No one checks their own work. The person who built or owns a model can't be the one who signs off on its validation — the system won't allow it.
Works with the major rulebooks banks and insurers answer to, including SR 26-2 (the US regulators' rulebook for checking that bank models work) and Annex IV (the EU AI Act's list of what technical documents an AI system must have).