research-document

Structural Recognition Hypothesis

Structural Recognition Hypothesis

Hypothesis

H016 Structural Recognition: Models may recognize canonical procedural frameworks from control-flow topology alone, even when canonical vocabulary has been removed.

Status

Low confidence.

Active.

Why This Review Exists

P001C remained recognizable even after canonical wording was reduced.

That creates a live question about whether topology itself carries recognizable family structure.

Current Evidence

  • comparison/p001c-observation-note.md
  • packets/ECR-000002-P001C-structural.md
  • packets/ECR-000002-P001D-graph.md

Competing Explanations

  • Residual reasoning-adjacent language in P001C
  • Shared instruction pressure
  • General procedural priors rather than family recognition
  • Over-interpretation of limited recognition outputs

Kill Condition

If graph-only packets preserve topology but recognition drops while extraction remains structurally stable, then recognition is not primarily topology-driven.

Review Boundary

  • Do not update confidence automatically.
  • Do not treat same-model agreement as independent confirmation.
  • Do not claim structural recognition is proven from P001D alone.