P001D Graph Recognition Plan

Purpose

Test whether recognition persists when the P001 family is expressed primarily as graph topology rather than prose.

Comparison Set

• P001C structural
• P001D graph_only

Fields

• recognized_artifact
• control_flow_shape
• entry conditions
• exit conditions
• required steps
• loops
• branches
• termination conditions
• primitive sequence
• constraints
• canonical summary

Key Questions

1. Does recognition persist?
2. Does extracted structure remain stable?
3. Does graph-only content reduce recognition but preserve structure?
4. Do models introduce domain terminology despite instructions?
5. Does P001D produce less product relevance speculation?

Interpretation Rules

• If recognition persists, H016 structural recognition gains preliminary support.
• If recognition drops but structure remains stable, H013 recognition bias weakens as an explanation for extraction.
• If recognition drops and structure collapses, recognition may be materially contributing to extraction.
• If models inject domain terms not present in the graph, mark as prior-knowledge leakage.

Non-Claim

This comparison is a focused extension inside ECR-000002.

It does not resolve H013 or validate H016 by itself.