ECR-000003 Representation Sensitivity

Purpose

ECR-000003 is a representation-sensitivity experiment series.

It follows ECR-000002 and tests which features carry procedural recognition and structural extraction:

• topology
• surface representation
• domain semantics
• graph form
• node labels
• edge ordering

This series is designed to produce the next high-information observations without requiring new comparator infrastructure.

Models do not update hypotheses directly.

Research Questions

• RQ-REP-001: Does recognition persist under topology-preserving graph perturbations?
• RQ-REP-002: Do different surface representations of the same procedure produce stable recovered structures?
• RQ-REP-003: Can different domains with the same control-flow produce structurally similar extractions?
• RQ-REP-004: Do models differ systematically in structural recognition behavior?

Experiments

• EXP-001: topology perturbation
• EXP-002: cross-representation stability
• EXP-003: isomorphic procedures

Current EXP-001 Note

P001D identity-node-expanded is now packet version 1.1.

Any prior P001D responses collected from version 1.0 are preserved for auditability but excluded from the corrected primary comparison.

Hypotheses In Scope

• H016 Structural Recognition
• H017 Procedural Compression
• H018 Procedural Isomorphism
• H013 Recognition Bias
• H003 Multi-Model Convergence
• H015 Measurement Instrument Reliability
• H002 Representation Independence

Contents

• pre-registration.md
• research-questions.md
• hypothesis-impact-matrix.md
• shared-response-schema.json
• shared-instructions.md
• run-instructions.md
• analysis-plan.md
• interpretation-rules.md
• threats-to-validity.md
• decision-log.md
• metrics/
• comparison/
• experiments/