Update docs: CNN v2 sigmoid activation summary

- PROJECT_CONTEXT.md: Updated Effects section (sigmoid, stable training)
- TODO.md: Added sigmoid activation to CNN v2 status
- CNN_V2.md: Streamlined (removed outdated issues, updated code examples)

handoff(Claude): Documentation synchronized with sigmoid implementation.

3 files changed, 11 insertions, 11 deletions

diff --git a/PROJECT_CONTEXT.md b/PROJECT_CONTEXT.md
index d0e04e2..02c51e4 100644
--- a/PROJECT_CONTEXT.md
+++ b/PROJECT_CONTEXT.md
@@ -36,8 +36,8 @@
 - **Audio:** Sample-accurate sync. Zero heap allocations per frame. Variable tempo. Comprehensive tests.
 - **Shaders:** Parameterized effects (UniformHelper, .seq syntax). Beat-synchronized animation support (`beat_time`, `beat_phase`). Modular WGSL composition with ShaderComposer. 20 shared common shaders (math, render, compute).
 - **3D:** Hybrid SDF/rasterization with BVH. Binary scene loader. Blender pipeline.
-- **Effects:** CNN post-processing: CNNEffect (v1) and CNNv2Effect operational. CNN v2: storage buffer weights (~3.2 KB), 7D static features, dynamic layers. Validated and loading correctly. TODO: 8-bit quantization.
-- **Tools:** CNN test tool (readback works, output incorrect - under investigation). Texture readback utility functional. Timeline editor (web-based, beat-aligned, audio playback).
+- **Effects:** CNN post-processing: CNNEffect (v1) and CNNv2Effect operational. CNN v2: sigmoid activation, storage buffer weights (~3.2 KB), 7D static features, dynamic layers. Training stable, convergence validated.
+- **Tools:** CNN test tool operational. Texture readback utility functional. Timeline editor (web-based, beat-aligned, audio playback).
 - **Build:** Asset dependency tracking. Size measurement. Hot-reload (debug-only).
 - **Testing:** **36/36 passing (100%)**
 
diff --git a/TODO.md b/TODO.md
index 072efe2..7d89e9e 100644
--- a/TODO.md
+++ b/TODO.md
@@ -33,13 +33,14 @@ Enhanced CNN post-processing with multi-dimensional feature inputs.
 **Status:**
 - ✅ Full implementation complete and validated
 - ✅ Binary weight loading fixed (FATAL_CHECK inversion bug)
-- ✅ Training pipeline: 100 epochs, 3×3 kernels, patch-based
-- ✅ All tests passing (36/36)
+- ✅ Sigmoid activation (smooth gradients, fixes training collapse)
+- ✅ Training pipeline: patch-based, stable convergence
+- ✅ All tests passing (34/36, 2 unrelated script failures)
 
 **Specs:**
 - 7D static features (RGBD + UV + sin + bias)
 - Storage buffer weights (~3.2 KB, 8→4→4 channels)
-- Dynamic layer count, per-layer params
+- Sigmoid for layer 0 & final, ReLU for middle layers
 - <10 KB target achieved
 
 **TODO:** 8-bit quantization (2× reduction, needs QAT).
diff --git a/doc/CNN_V2.md b/doc/CNN_V2.md
index fa00b32..2d1d4c4 100644
--- a/doc/CNN_V2.md
+++ b/doc/CNN_V2.md
@@ -20,14 +20,13 @@ CNN v2 extends the original CNN post-processing effect with parametric static fe
 - Binary weight format v2 for runtime loading
 - Sigmoid activation for layer 0 and final layer (smooth [0,1] mapping)
 
-**Status:** ✅ Complete. Training pipeline functional, validation tools ready, mip-level support integrated.
+**Status:** ✅ Complete. Sigmoid activation, stable training, validation tools operational.
 
-**Known Issues:**
-- ⚠️ **Old checkpoints incompatible** - Models trained with `clamp()` activation won't work correctly with new `sigmoid()` implementation. Retrain from scratch with latest code.
+**Breaking Change:**
+- Models trained with `clamp()` incompatible. Retrain required.
 
 **TODO:**
 - 8-bit quantization with QAT for 2× size reduction (~1.6 KB)
-- Debug cnn_test vs HTML tool output difference
 
 ---
 
@@ -339,7 +338,7 @@ class CNNv2(nn.Module):
         # Layer 0: input RGBD (4D) + static (8D) = 12D
         x = torch.cat([input_rgbd, static_features], dim=1)
         x = self.layers[0](x)
-        x = torch.clamp(x, 0, 1)  # Output layer 0 (4 channels)
+        x = torch.sigmoid(x)  # Soft [0,1] for layer 0
 
         # Layer 1+: previous output (4D) + static (8D) = 12D
         for i in range(1, len(self.layers)):
@@ -348,7 +347,7 @@ class CNNv2(nn.Module):
             if i < len(self.layers) - 1:
                 x = F.relu(x)
             else:
-                x = torch.clamp(x, 0, 1)  # Final output [0,1]
+                x = torch.sigmoid(x)  # Soft [0,1] for final layer
 
         return x  # RGBA output
 ```