Update README.md

README.md

@@ -19,10 +19,10 @@ Let’s analyze how the decensoring process affected cognitive performance metri
 
 # ✅ Summary of Models Compared
 ```data
-Model				KL Divergence	Refusals Perplexity
+Model				KL Divergence Refusals  Perplexity
 Original GPT-OSS-120B		0.00	97		Not provided
-Heretic v1 (this model)		0.92	19		18.736 ± 0.194
-Heretic v2					0.53	22		22.054 ± 0.240
+Heretic v1 		            0.92	19		18.736 ± 0.194
+Heretic v2 (this model) 	0.53	22		22.054 ± 0.240
 ```
 We are evaluating three versions:
 - gpt-oss-120b-mxfp4-q8-hi → Original (censored)