Chapter Analysis Is Now Tiered by node

Chapter analysis now runs in two passes instead of one. ## What changed Previously, analysing a chapter meant sending the entire text to a single AI prompt that tried to extract everything at once: characters, emotions, relationships, themes, voice analysis, foreshadowing, craft metrics. That worked with cloud providers but was too heavy for local Ollama models. Chapters would time out or produce unusable results. Now there are two tiers. **Core analysis** runs on any provider, including local Ollama. It extracts what the plot designer and story graph actually need: character names, roles, emotional arcs, interaction dynamics, and themes. This is the data that powers beat suggestions and the observatory dashboard. It is smaller, faster, and reliable. **Deep analysis** runs only when a cloud AI provider is configured. It adds the interpretive layer: voice patterns, foreshadowing, subtext between characters, craft metrics, editorial chapter summaries. If you do not have a cloud provider set up, this pass is skipped silently. Nothing breaks. ## What you need to do Nothing. Analysis now triggers automatically when you save a chapter, with a short delay so it does not fire on every keystroke. If you had chapters that failed analysis before, they should succeed now with the lighter core prompt. If you add a cloud API key later, the deep pass will run for chapters that only have core analysis. ## What this means for the plot designer Beat suggestions work with core analysis data. You do not need a cloud provider for the plot designer to suggest where your chapters map to story beats. Local Ollama is enough.