docs/PHASE2_UI_REPORT.md

Phase 2 UI Report

Scope

Phase 2 polishes the mock MVP into an archive-style Gradio interface while preserving the current deterministic pipeline.

Completed:

• archive dashboard layout: Object Intake, Object File, Secret Diary, Share Card, Object Chat, and Trace
• English-first / Chinese-second UI hierarchy
• dark paper archive styling with amber highlights and museum-label panels
• segmented personality mode control
• six visible example object cards that trigger stable mock generation
• empty states for Object File, Diary, Share Card, and Trace
• UI error fallback that exposes the exception type and message without hiding generation context
• screenshot-friendly share card rendering
• mobile single-column layout for narrow screens

Not included:

• real MiniCPM-V or other VLM integration
• real llama.cpp / llama-cpp-python text runtime
• LoRA fine-tuning
• Hugging Face Space deployment
• GitHub remote creation
• external image assets or production dependencies

Reference Boundary

The local UI 参考/ images were used only for visual direction: archive panels, dark paper texture, amber linework, example cards, and museum-label hierarchy.

UI 参考/ remains ignored by Git and must not be committed.

Runtime Boundary

The app still runs on the mock runtime. The UI calls the existing src/pipeline.py boundary and does not change model behavior.

Verification Targets

• desktop preview at 1440x1000
• mobile preview at 390x844
• six example object cards generate object JSON, persona JSON, diary, share card, and trace
• manual description plus personality mode still generates successfully
• chat shows "Wake an object first. / 请先唤醒一个物品。" before generation