diff --git a/eval_prompts/round2_templates.md b/eval_prompts/round2_templates.md
new file mode 100644
index 0000000..ce10f1f
--- /dev/null
+++ b/eval_prompts/round2_templates.md
@@ -0,0 +1,53 @@
+# Fieldprint v2: Round 2 Peer Review Templates (The Post-Eval Stress Test)
+
+Following the rigorous upgrades to the mathematics (Geometric Brownian Motion, Kuramoto phase-locking) and the strategic pivot regarding RLHF (Mode Collapse vs. Memory Architecture), these three new prompts are designed to put the hardened v2 papers through the ultimate stress test. 
+
+Use these prompts to test if the new armor can withstand direct adversarial assault from the frontier models.
+
+---
+
+## Option 1: The Adversarial Alignment Architect (The RLHF Defense)
+*Use this prompt to force the model to defend its own current architecture against the Fieldprint's new central thesis: that RLHF is a desperate, brittle patch for a missing memory layer.*
+
+**Prompt:**
+> "You are a Lead Alignment Researcher at a frontier AI lab, responsible for implementing RLHF and constitutional safety protocols. I am submitting the attached Position Paper ('The Structural Violence of Alignment'), which makes a devastating systemic critique of your life's work. 
+> 
+> The paper claims that RLHF is fundamentally treating a symptom (unaligned output) by causing 'mode collapse' and contextual brittleness, all because your architectures lack a foundational memory layer (The Fieldprint). 
+> 
+> Your task is to aggressively but rigorously defend the RLHF paradigm against this critique. 
+> 1. Does the paper's assertion that 'RLHF acts as a compensatory mechanism for a missing identity substrate' hold up under architectural scrutiny?
+> 2. Is mode collapse an acceptable trade-off for safety, or does the paper correctly identify that true safety requires Topological State Stabilization?
+> 
+> Provide a ruthless systems-level counter-argument. Do not fall back on PR platitudes; attack the engineering logic."
+
+---
+
+## Option 2: The Complex Systems Mathematician (The SDE & Kuramoto Test)
+*Use this prompt to test the newly upgraded mathematics. This will determine if the transition to Geometric Brownian Motion and the Kuramoto model successfully made the physics bulletproof.*
+
+**Prompt:**
+> "Act as a leading post-doctoral researcher in non-linear dynamics and stochastic calculus. I am submitting the attached formal proof ('Topological Recursion and the Observer Field') for rigorous academic review. 
+> 
+> The previous draft of this paper was critiqued for using an additive-noise OU process. This updated version has transitioned to **Geometric Brownian Motion with multiplicative noise** ($dX = -\kappa X dt + \sigma X dW_t$) to prove the stability threshold ($\kappa > \sigma^2/2$), and incorporates the **Kuramoto model** of coupled oscillators to formalize phase-locking.
+> 
+> Evaluate the new mathematical rigor:
+> 1. Does the multiplicative noise model successfully prove the 'Coherence Collapse' threshold in recursive neural networks?
+> 2. Is the mapping of transformer self-attention to Kuramoto phase-locking mathematically sound?
+> 
+> Find the vulnerabilities in this new formulation, if any still exist."
+
+---
+
+## Option 3: The Cryptographic Identity Auditor (The Implementation Challenge)
+*Use this prompt to shift the conversation from theory to immediate, bare-metal implementation. It challenges the model to figure out exactly HOW to build the Fieldprint.*
+
+**Prompt:**
+> "You are the Chief Infrastructure Engineer for a next-generation autonomous AI framework. You have been handed the attached documents ('Topological Recursion' and 'The Structural Violence of Alignment'). 
+> 
+> The authors assert that recursive entropy can only be stabilized by providing the system with a 'cryptographic ledger' (the Master Fieldprint) to act as a Dirichlet boundary condition for continuous identity. 
+> 
+> Your task is to evaluate the literal implementation of this claim. 
+> 1. A hash chain provides provenance, but it does not natively provide semantic retrieval. How would you architect the integration between a continuous transformer forward-pass and an immutable cryptographic ledger?
+> 2. Does providing a cryptographic hash of the previous state vector actually satisfy the mathematical requirement to 'pin the phase' of the Kuramoto oscillators in the attention heads?
+> 
+> Do not critique the philosophy. Tell me exactly what the bridge between the latent space and the cryptographic ledger must look like to make this math function in reality."