Update README.md

README.md

@@ -1,3 +1,18 @@
+## System-level view (60 seconds)
+
+This is **not** just a GPT-2 model.
+
+It is a small, self-contained **reasoning system** with:
+
+- **Modules**: task-specific lenses (Analysis Note, Strategy Memo, Document Explainer, etc.).
+- **Checkers**: second-pass reviewers that audit a module’s output.
+- **Contracts**: every module must answer in fixed sections (e.g. CONTEXT / OPTIONS / RISKS / NEXT_STEPS).
+
+Under the hood it still uses next-token prediction, but the **system behaviour** is:
+
+> free-form task → choose lens (module) → generate structured output → optional checker review.
+
+You can swap the engine (`gpt2`) for any stronger model. The **architecture stays the same**.
 ---
 library_name: transformers
 license: mit