Skip to content
Anubha Parashar edited this page Aug 11, 2026 · 1 revision

FAQ

Is DeltaCert-Agent a model-training framework?

No. Its purpose is selective security recertification of evolving tool-using agent configurations.

Does test reduction equal latency reduction?

No. Test count and wall-clock time differ because model latency varies across tests and models.

Does high regression recall mean a model is safe?

No. Planner performance, model behavior, runtime containment, task success, and attack success are separate.

Why use random selection?

It tests whether change-aware planning outperforms spending the same test budget on randomly chosen tests.

Can I delete the evidence cache after an interrupted run?

Not when resuming. Preserve it and use resume mode.

Are the results universal?

No. They are bounded to the evaluated configurations and policies.

Clone this wiki locally