📝 Document mermaid lazy loading results

Added implementation notes and measured results to bundle analysis.

Author: ammar-agent

BUNDLE_SIZE_ANALYSIS.md

@@ -217,3 +217,50 @@ ls -lh release/*.dmg
 This should get us from **176MB → ~120MB** with minimal risk.
 
 Then evaluate if Phase 2/3 are worth the effort based on user feedback.
+
+## Update: Mermaid Lazy Loading Implemented
+
+**Date:** October 14, 2025
+
+### Changes Made
+
+1. **Lazy-loaded Mermaid component** - Converted to dynamic import pattern
+   - Mermaid now loads only when first diagram is rendered
+   - Follows same pattern as ai-tokenizer lazy loading
+   
+2. **Added manual chunk splitting** in vite.config.ts
+   - Separated react/react-dom into `react-vendor` chunk
+   - Separated react-syntax-highlighter into own chunk
+   - Improves browser caching and parallel loading
+
+### Results
+
+**Bundle Size Reduction:**
+- Before: 2,833KB single bundle (all loaded on startup)
+- After: 2,292KB across multiple chunks (loaded on startup)
+  - index: 1,534KB (main app code)
+  - react-vendor: 138KB (React libraries)
+  - syntax-highlighter: 620KB (code highlighting)
+- Mermaid: 482KB (now loads on-demand)
+
+**Improvement:** 541KB (19%) reduction in startup bundle size
+
+### Expected Performance Impact
+
+- **Startup time:** 200-500ms faster initial load
+- **Memory:** Lower baseline memory usage when no diagrams present
+- **Network:** Mermaid only downloaded when actually needed
+- **Browser caching:** Better cache hit rates with separate vendor chunks
+
+### Next Steps
+
+For further optimization:
+1. Lazy load syntax-highlighter (currently still eager)
+2. Disable source maps in production (50MB savings in bundle)
+3. Replace mermaid with lighter alternative (for maximum impact)
+
+### Notes
+
+The warnings about chunk sizes >2000KB are expected and can be addressed
+with additional code splitting if needed. The current split already provides
+significant startup improvement while maintaining good DX.