# send_dynamic_card Complete Flow Diagram
> **Note:** force_reindex_components
## Complete Mermaid Diagram
```mermaid
flowchart TD
subgraph MCP["🔧 MCP Tool: send_dynamic_card"]
PARAMS["Parameters:<br/>• user_google_email*<br/>• space_id*<br/>• card_description*<br/>• card_params<br/>• thread_key<br/>• webhook_url"]
end
PARAMS --> SCB["SmartCardBuilder<br/>build_card_from_description()"]
subgraph SMART["SmartCardBuilder (Primary Path)"]
SCB --> STRICT{strict_mode?}
STRICT -->|Yes| EXPLICIT["Build from<br/>explicit params only"]
STRICT -->|No| GRID{grid or<br/>images?}
GRID -->|Yes| GRID_CARD["_build_grid_card()"]
GRID -->|No| SECTIONS{explicit<br/>sections?}
SECTIONS -->|Yes| SECTION_CARD["_build_card_from_explicit_sections()"]
SECTIONS -->|No| FEEDBACK["_get_proven_params()"]
subgraph QDRANT1["🔍 Qdrant: Feedback Loop"]
FEEDBACK --> Q1["Query mcp_gchat_cards_v7<br/>for proven patterns"]
end
FEEDBACK --> PATTERNS["_find_similar_patterns()"]
subgraph QDRANT2["🔍 Qdrant: Pattern Search"]
PATTERNS --> Q2["_search_by_identity()<br/>ColBERT → 'components' vector"]
Q2 --> Q2_ENRICH["Enrich with full payload:<br/>instance_params, parent_paths"]
end
PATTERNS --> STRUCT["StructureValidator<br/>generate_structure_from_inputs()"]
STRUCT --> TEMPLATE["_find_matching_template()"]
subgraph QDRANT3["🔍 Qdrant: Template Search"]
TEMPLATE --> Q3["Query promoted templates<br/>'relationships' vector"]
end
TEMPLATE --> NL_PARSE["Parse NL description<br/>into sections/widgets"]
NL_PARSE --> RENDER["_render_card()"]
subgraph QDRANT4["🔍 Qdrant: Component Loading"]
RENDER --> Q4["_search_by_relationships()<br/>MiniLM → 'relationships' vector<br/>(symbol-enriched embeddings)"]
Q4 --> LOAD["ModuleWrapper.get_component_by_path()"]
end
end
EXPLICIT --> CARD_OUT["Card JSON"]
GRID_CARD --> CARD_OUT
SECTION_CARD --> CARD_OUT
RENDER --> CARD_OUT
CARD_OUT --> CHECK{Card<br/>created?}
CHECK -->|No| FALLBACK["_build_simple_card_structure()<br/>(No Qdrant - basic card)"]
CHECK -->|Yes| SEND["Send to Google Chat"]
FALLBACK --> SEND
SEND --> API{webhook_url?}
API -->|Yes| WEBHOOK["POST to Webhook"]
API -->|No| GCHAT_API["Google Chat API<br/>spaces.messages.create()"]
WEBHOOK --> RESPONSE["SendDynamicCardResponse"]
GCHAT_API --> RESPONSE
style QDRANT1 fill:#e1f5fe,stroke:#01579b
style QDRANT2 fill:#e1f5fe,stroke:#01579b
style QDRANT3 fill:#e1f5fe,stroke:#01579b
style QDRANT4 fill:#e1f5fe,stroke:#01579b
style FALLBACK fill:#fff3e0,stroke:#e65100
```
## Key Points
### All primary paths use Qdrant:
1. **Feedback Loop** → queries proven patterns from successful cards
2. **Pattern Search** → `_search_by_identity()` with ColBERT multi-vector on `components` named vector
3. **Template Search** → queries promoted templates via `relationships` vector
4. **Component Loading** → `_search_by_relationships()` with MiniLM on `relationships` vector (this is where symbol-enriched embeddings match DSL notation)
### Only fallback bypasses Qdrant:
- `_build_simple_card_structure()` is only used if SmartCardBuilder throws an exception - it builds a basic text/button card without any vector search
### DSL symbols work via vector similarity:
- The v7 collection's `relationships` vector contains symbol-enriched text like:
- `§ Section | contains C_0 CollapseControl, ʍ Widget | §[C_0, ʍ]`
- `Ƀ ButtonList | contains ᵬ Button | Ƀ[ᵬ]`
- When you pass `§[δ, Ƀ[ᵬ×2]] Dashboard` as `card_description`, the MiniLM embedding matches against these symbol patterns
---
✻ *Cooked for 1m 9s*
