Numbers

Live counts from this specific deployment's database — real gate decisions and orders, not a benchmark. The controlled comparison (200 seeded sessions, Arm A naive vs Arm B PRAMAN, including false-positive cost, reversibility-band accuracy, and injection-invariance) is a separate, already-completed run in harness/RESULTS.md, kept distinct from this page rather than blended into it — a live deployment counter and a controlled benchmark answer different questions, and conflating them would make both less trustworthy.