# Lesson 04 — The high speed trader (advanced-medium) > Series: [index](README.md) · prev: [03](03-counsel-choosing-an-approach.md) · next: [05 — Pragmatics and evolution](05-pragmatics-and-evolution.md) Module: `lib/advantage/modules/trader` (see the [modules README](../../lib/advantage/modules/README.md)). Goal: at least **5,000 evaluations per second** and about **2,000 trades per minute** (roughly 0.7% of candidates), on synthetic data. ## What you do 1. Run it: `sleela run HstDemo.sleela` from `lib/advantage/modules/trader`, or load it as a module (lesson 09). 2. Read section 2 of the output: evals/s, trades per simulated minute, vetoes by phase. 3. Find where 99% of candidates die. Most are score rejects and logic vetoes. Why is that the right place for them to die? ## Why the numbers fit together 5,000 evaluations per second at 2,000 trades per minute means about 33 trades per second, so the trader must say **no** about 150 times for every yes. Speed is mostly cheap rejection: hard gates first, scoring only for survivors, and a token bucket so a burst of good candidates cannot exceed the cap. ## Exercises - Change the target to 600 trades per minute. What happens to the threshold? - Raise `evalsPerSimSecond` to 20,000. Does the model keep up in wall time? - Why does P = 24 cost more per evaluation than P = 8? ## Advanced Make the veto path faster: reorder so the most frequent veto comes first. Measure.