Notes on classification economics, intent, agentic risk, and the attacks we catch in real corpora. Written to be read, and shared the day it goes live.
✓ Thanks — you're on the list.
Why a four-layer cascade beats brute-forcing every document through a frontier model, on both accuracy and cost.
A single prompt looks benign. Read the lineage across a session and the campaign gives itself away.
A tainted document, a helpful agent, and the plan-level check that caught the exfiltration before it ran.
Benchmarks, intent research, and the attacks we catch in real corpora. No noise, just what we find while building.
✓ Thanks — you're on the list.