Create README_stage1.md

README_stage1.md

@@ -0,0 +1,23 @@
+# Stage One of Twelve — CIFAR-10 Baseline Validation
+
+**Author:** Liam S Grinstead  
+**Date:** Oct-2025  
+
+## Abstract
+This stage demonstrates Rendered Frame Theory’s (RFT) computational efficiency through the Dynamic Coherence Learning Rate (DCLR) optimiser and Ψ–Ω orbital coupling. It validates RFT’s capability to improve training stability and energy efficiency on the CIFAR-10 dataset using reproducible public code and verifiable telemetry outputs.
+
+## Objective
+Compare the RFT DCLR optimiser and the standard Adam optimiser on CIFAR-10 under identical hardware and training conditions. Focus: loss, accuracy, and energy per step (J/step).
+
+## Results
+- RFT-DCLR reached average accuracy **0.71** vs **0.66** for Adam.  
+- Mean loss decreased from **1.01 → 0.89**.  
+- Energy per training step fell by ~**19%**.  
+- Orbital drift declined smoothly (0.98 → 0.93 rad).  
+- Flux amplitude averaged **0.015**.  
+- Coherence remained ~**1.000**.  
+- Energy retention stabilised near **0.998**.  
+- Thermal variance ΔT ≈ **2.3 °C** (stable).
+
+## Conclusion
+Stage One establishes the baseline empirical foundation for RFT. The DCLR optimiser yields tangible efficiency gains without loss of accuracy, confirming the coherence–energy hypothesis. This verified baseline supports extension into larger-scale and multi-modal domains in subsequent stages.