What This Chart Shows
This chart visualizes how raw business data flows through the system—from initial data ingestion, through GPU computation, to the final attribution outputs and pattern intelligence. Each node represents a data family (a specific type of stored data), and node size indicates how much GPU memory (VRAM) that family consumes. This is primarily an architecture and diagnostic view, helping technical users and analysts understand the system's data provenance and memory footprint.
Key Questions This Chart Helps Answer
- What data families power the attribution and pattern intelligence outputs?
- How does data flow from raw events to final attribution results?
- Which data families consume the most GPU memory?
Axes, Metrics, and Units
Element | Description |
|---|---|
| Layout | Vertical DAG (directed acyclic graph) in org-chart style. Top = inputs, Bottom = outputs. |
| Node size | Proportional to GPU memory (VRAM) usage in Gigabytes (GB). Larger nodes = more memory. |
| Node color | Bridge type: Blue = Global Bridge (inputs) Purple = CSR Bridge (attribution) Orange = Compute Engine Green = Edge Bridge (patterns). |
| Node labels | Data family name with internal tensor name in italics. |
| Edges | Data flow relationships from parent to child families. Bracket-style routing for clarity. |
| Hover | Shows full family name, bridge type, VRAM size in GB, and description. |
All size values are measured from actual GPU memory allocation.
Control Options Reference
This chart has no user-configurable controls. It reflects the current state of the data pipeline as loaded in memory.
How to Interpret the Results
- Top row (Generation 1 - Ancestors): Raw inputs: business events and enriched metadata from the Global bridge.
- Second row (Generation 2 - Parents): Core structures: graph topology, marketing schedules, and temporal windows on the CSR bridge.
- Third row (Generation 3 - Engine): The GPU Compute Engine (Dual-Track + gSpan) that processes all data.
- Bottom row (Generation 4 - Outputs): Two output branches:
- Left (CSR Bridge): Attribution ledgers—daily attribution and per-lever ROI.
- Right (Edge Bridge): Pattern intelligence—pattern summaries, path provenance, and UI reverse index.
- Node size: Larger nodes consume more GPU memory. The graph topology and path provenance are typically the largest.
- Missing nodes: If a node is unusually small or missing, that data family may not be fully loaded.
Practical Applications for Marketers
| Application | How to use this chart |
|---|---|
| System health check | Verify that all expected data families are loaded and sized appropriately. |
| Memory troubleshooting | Identify which families are consuming the most VRAM if performance issues arise. |
| Data provenance | Understand where attribution and pattern outputs come from in the pipeline. |
| Onboarding / training | Use as a reference diagram to explain how the platform processes marketing data. |
| Technical audits | Demonstrate data lineage and architecture to technical stakeholders. |
Common Mistakes & Misinterpretations
| Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating this as a marketing performance chart | This is an architecture/diagnostic view, not a performance metric. It shows system structure, not campaign results. | Use this for system understanding, not for marketing decisions. |
| Assuming node size = importance | Node size = memory usage, not business importance. A small node can be critical; a large node may be intermediate data. | Interpret size as a technical metric (memory), not a business metric. |
| Expecting interactivity | This chart is primarily informational. Nodes do not drill into detailed metrics. | Use other charts (Attribution Waterfall, etc.) for actionable marketing insights. |
| Ignoring missing families | If expected families are missing or very small, data may not be fully loaded. | Check hover tooltips for actual sizes; consult technical support if families are unexpectedly absent. |
| Comparing sizes across different runs | Memory sizes can vary based on data volume and configuration. | Compare within the same run; avoid cross-run size comparisons without context. |
Caveats & Considerations
- Technical audience: This chart is most useful for technical users, data engineers, and analysts who need to understand the system architecture. Marketers may find other charts more actionable.
- Memory sizes: Sizes are measured in GiB but displayed as GB for readability. These are actual GPU memory allocations, not theoretical maximums.
- Fallback values: If a data family is not loaded, a fallback size is used to maintain visual layout. Check hover for actual vs fallback.
- 6-family provenance: The chart reflects the newer "6-family provenance contract" which splits outputs into attribution (CSR) and pattern intelligence (Edge) branches.
- Pipeline completeness: A fully loaded pipeline should show all nodes with reasonable sizes. Missing or zero-sized nodes may indicate incomplete data ingestion.