**L16 Prompt Drift: GPT-2 vs LLaMA-3-8B — IDENTICAL DRIFT**

Just ran the same L16 on **LLaMA-3-8B-Instruct**.

**Result**:

- **Exact same drift profile**

- **Certainty + Rare Word = +0.70 drift (p<0.01)**

- **Truth = 1.0 in all 32 runs**

**Live Colab (switch models in 1 line)**:

https://colab.research.google.com/drive/1CPUu9LhE-fBAwrsSA2z53hufIDsf1ed_?usp=sharing

**This is not model-specific. This is prompt physics.**

Next: Mistral, Gemma, Claude