Align docs around LLM-driven scenario generation

docs/demo-script.md

@@ -25,7 +25,7 @@ Show the app working end to end so judges can evaluate it even if API limits, mo
    `Mentor`
 
 6. Click `Enter this future`.
-   Point out that the first scene starts immediately.
+   Point out that the app is built around a 7B model for adaptive future scenarios, with deterministic state logic keeping the simulation grounded.
 
 7. Show the first scenario card, source label, metric panel, persona bubble, world state, and character sprite.
 
@@ -44,7 +44,7 @@ This is LifeChoice Simulator, a causal future simulator for difficult decisions.
 
 Instead of giving one-shot advice, the app lets me enter a path and experience a short sequence of consequences. I choose MTech over a Software Job, add the real pressure behind the decision, and pick a mentor voice.
 
-The first scene loads immediately because it is deterministic. Later scenes can be enriched by a 7B model, but the simulation engine owns the actual state: metrics, facts, obligations, closed options, safety rules, and the final report.
+The adaptive scenario experience is powered by a 7B model. The simulation engine owns the actual state around it: metrics, facts, obligations, closed options, safety rules, validation, and the final report.
 
 Every choice changes the world. The metric panel updates, the environment can move between stable, thriving, and struggling, and the character can become neutral, stressed, or confident.