How X-Ray and the Causal Graph Work

Every line you read was produced by a step. Every step belongs to a phase. Every phase belongs to a session. This is not metadata bolted on after the fact. It is structural. It is the system's memory.

X-Ray: Span-Level Attribution

X-Ray tracks who wrote what. At the line level, every line is assigned a collaboration type: ai_only, human_only, or collaborative. Blue tinting means AI originated the line. Green means a human did. Warm amber means both contributed.

But X-Ray goes deeper. When a line has mixed authorship, X-Ray resolves attribution down to the span level. Individual words and phrases within a single line are tinted independently, showing exactly where AI ended and the human began. Hover any tinted span to see its origin step, last editor, and collaboration type.

This is the difference between claiming provenance and proving it. Line-level shows who owns a line. Span-level shows who owns every word.

The Causal Graph: Hierarchical Drill-Down

The Causal Graph is not a flat diagram. It is a three-level hierarchy designed for three audiences.

At the top level, Sessions present the executive view. Large cards, generous whitespace, chronological flow. Each session shows its time range and a summary badge counting phases and steps. Cross-session influence edges are lifted from the step level and displayed as summary connections.

Click a session to drill into Phases. This is the architect's view, optimized for readability. Phases display their goal, step count, and actor breakdown. Inter-phase edges are lifted from step-level connections, aggressively deduplicated, so the flow reads clean. One clear end-to-end pipeline.

Click a phase to reach Steps. This is the developer's view. Every step shows its actor dot, and raw production and influence edges are drawn directly. Cross-phase connections appear as ghost nodes with dashed borders and italic labels. Click any step to open a flyout showing timestamp, duration, output summary, and a clickable influence chain.

The graph above shows how this blog post was produced. Session 1 was research and drafting. Session 2 was editing and visual insertion. Breadcrumb navigation lets you move back up at any level. The layout engine adapts spacing to node density, keeping small graphs spacious and large graphs readable.

Putting It Together

Switch between Document, X-Ray, and Causal Graph views using the toggle bar above. Each view answers a different question:

- Document: What does the content say? - X-Ray: Who wrote each word? - Causal Graph: How was the content produced?

This is not decoration. This is governance infrastructure. The system does not guess. It traces.