Update README.md

README.md

@@ -58,6 +58,10 @@ This model represents a fundamental shift in how we approach LLM fine-tuning. In
 
 ---
 
+### 🏆 新流水线：The "Iron Logic" Pipeline
+
+`Base Model` -> **`FT (Logic Injection)`** -> `Standard SFT (Style Polish)` -> `RLVR (Reasoning)`
+
 # 🏋️ Fragmented Training: The "Cognitive Burden" Paradigm
 ### A Novel Approach for Accelerated & Enhanced Logic in LLMs
 
@@ -65,10 +69,6 @@ This model represents a fundamental shift in how we approach LLM fine-tuning. In
 **Base Model:** Qwen3-4B (Thinking-2507)
 **Methodology:** Stochastic Token Shuffling (70% Noise Rate)
 
-### 🏆 新流水线：The "Iron Logic" Pipeline
-
-`Base Model` -> **`FT (Logic Injection)`** -> `Standard SFT (Style Polish)` -> `RLVR (Reasoning)`
-
 ---
 
 ## ⚡ Key Results at a Glance