# Lesson 09 — The medium and the bridge (advanced) > Series: [index](README.md) · prev: [08](08-tall-science.md) The AI modules are Advantage™ components. A **user defined program** loads the **medium** (the center); the medium holds modules and owns the **bridge** that carries messages in both directions. Reference: [modules README](../../lib/advantage/modules/README.md). Example program: [user-program/UserProgram.sleela](user-program/UserProgram.sleela). ## Read it first Run `UserProgram.sleela`, then read the order of events: the program loads only the modules it wants; modules announce `ready` on their own; the program asks and gets replies; the evaluator volunteers a `cardinal-veto` event nobody asked for. ## Three directions 1. Program to module: `ask` waits for the reply; `tell` does not. 2. Module to program: replies plus unsolicited events, handed to your listener on `pump`. 3. Module to module: `link` the trader to the evaluator and each accepted trade is judged; a NO withdraws it. The trader reports `trade-judged` upward. ## Safety you should be able to explain Bounded queues (refuse, count, never grow); reserved names; error replies for unknown modules and topics; a 100,000-message pump cap against ping-pong; a closed bridge after `shutdown`. Why is refusing better than blocking here? ## Exercises 1. Write your own module (implement `SLAdvModule`) that counts every `trade-judged` event. How does it hear them if events go to `program`? 2. Load only the evaluator and judge 1,000 trades. What does the ledger say? 3. Make two of your modules answer each other and watch `wasSaturated()`. 4. Moral verdicts for high speed trades are YES, never COMPLETE. Which two time points fail, and is that the right outcome?