refactor(physics): definitive mathematical rigorous fixes for Round 5 critiques

papers/Adversarial_Review_Logs.md

@@ -139,3 +139,35 @@
 ### 3. Turing Completeness in Continuous Time
 **Conflating Combinational Logic with Sequential Memory:** The constructed asynchronous "AND gate" was actually a Muller C-element (a sequential state machine). Furthermore, a single intermediate state $M$ cannot differentiate between asynchronous input orders.
 **The Fix:** Define distinct saddle states $M_A$ and $M_B$. Re-label the "AND gate" accurately as an asynchronous C-element or sequential join, defining the exact Lotka-Volterra inhibitory matrix for routing.
+
+---
+
+## Log 9: The Physicist's Critique (Round 5 - Core Rigor)
+
+### 1. Relativistic Latency in Markovian Networks
+**The Diffusion vs. Wave Error:** Markov processes yield diffusion (first-order in time), not relativistic wave propagation. A Lieb-Robinson speed limit is not Lorentz invariance (it preserves a lattice preferred frame).
+**The Fix:** Derive the Poincaré algebra directly from the continuum limit of the graph, showing how the metric tensor $g_{\mu\nu}$ emerges.
+
+### 2. Recursive Witness Dynamics and Quantum Darwinism
+**Ontological Contradictions:** Conflating classical stochastic Markov transition matrices with quantum Pauli spins. The claim that $I(S:E_f)$ proves redundant copies ignores that a specific initial environmental superposition is required for dephasing to occur.
+**The Fix:** Pick a strict quantum lattice Hamiltonian model. 
+
+### 3. Holographic Entanglement Entropy in Markovian Networks
+**Classical Thermalization vs. Quantum Evaporation:** Classical Markov leakage is thermalization, where relative entropy strictly decreases; it never drops to zero like the Page curve.
+**The Fix:** To achieve the Page curve, mathematically demonstrate a globally pure state where the tensor factor of the "interior" decreases in dimension over time.
+
+---
+
+## Log 10: The Logician's Critique (Round 5 - Core Rigor)
+
+### 1. The Intellecton as the Minimum Viable Markov Blanket
+**Ontological Misalignment & Ignored IIT:** A generating partition over a stochastic system is mathematically fraught. Mapping $P$ only to $S \to I$ orphans the external world $E$. Furthermore, a Markov blanket does not guarantee high Integrated Information ($\Phi$).
+**The Fix:** Define how the World $E$ maps through $S$ into $I$, and explicitly require Tononi's $\Phi > 0$.
+
+### 2. Rate-Distortion Theory in Markovian Networks
+**Causal Graph Suicide & DPI Violation:** Fitness is a *collider* ($X \to F \leftarrow A$). DPI fails at colliders. Using the Information Bottleneck method backwards.
+**The Fix:** Use standard Rate-Distortion Theory where distortion is $D(x, y) = -\max_a \mathbb{E}[F(x, a) \mid y]$. 
+
+### 3. Turing Completeness in Continuous Time
+**Saddles vs. Attractors & The One-Shot Fallacy:** A saddle cannot store memory indefinitely; noise will kick it out along its unstable manifold. A Muller C-element must also reset to be reusable.
+**The Fix:** Treat inputs as *bifurcation parameters* that alter stability. Explicitly define the reset cycle $C \to R$.