No Model.
Kestri works with the evidence itself.
Decode any model response.
Give Kestri the output. She recovers the evidence —
including what the model missed — and tests the answer
against the latest public and private data you can access.
Ask the locals. Compare the documents.
For compliance, Kestri retrieves current system-of-record
evidence and compares the governing document directly with
the document being tested. No training cutoff. No retraining.
No re-embedding.
A model can answer. Kestri shows where it says that.
