Switch back to Inference Providers (Qwen3.5-9B)

Run Qwen3.6-35B-A3B locally on ZeroGPU instead of Inference API

Use tuple[str,str] return annotation so Gradio counts 2 outputs

Annotate generate_bulletin return type so Gradio routes 2 outputs

Split frontend into custom HTML via gradio.Server

Update app.py

Initial bulletin app: Qwen3.6-35B-A3B → vintage 1080×1440 retro report