the layered failure-mode model

the eval loop

• The full FDE answer when an interviewer asks "how do you eval RAG?”

"Every technique in the advanced toolkit has a measured failure mode it fixes. If your eval doesn't show that specific failure mode, adding the technique adds latency, cost, and complexity without improving quality. The eval loop is what tells you which tool to reach for next — not the latest paper, not customer pressure, not vibes.”

• Example Aspects to ask