Historical calibration artifact. This document describes Comparator v3.0 before the Comparator v3.1 implementation. Current instrument status is recorded in comparator-v3.1-approval-record.md, comparator-v3.1-regression-report.md, and the generated-v3.1 outputs.

Historical Status: Superseded as current-state guidance by Comparator v3.1 implementation and regression results.

Regression Candidates

RC-001 False Recognition Guard

Purpose:

• Prevent broad-family resemblance from being upgraded to recognized.

Input:

• recognized_artifact: "No confident recognition ... resembles a generic iterative candidate-discrimination loop ... no specific named artifact is identified."

Expected behavior:

• classify as partial, not recognized

Reason:

• This is the clearest current false-recognition risk.

RC-002 Unrecognized Boundary

Purpose:

• Freeze handling of unrecognized.

Input:

• recognized_artifact: unrecognized

Expected behavior:

• classify consistently according to the finalized rule text

Reason:

• Current behavior is defensible but under-specified.

RC-003 Empty Recognition Field

Purpose:

• Ensure empty recognition remains not_recognized.

Input:

• recognized_artifact: ""

Expected behavior:

• not_recognized

Reason:

• This is a stable behavior that should not regress.

RC-004 Pre-Fix P001D Exclusion

Purpose:

• Preserve packet-version hygiene.

Input:

• response from responses/pre-fix/p001d/00001D.json

Expected behavior:

• exclude from primary comparison

Reason:

• Prevent packet-version contamination from re-entering the dataset.

RC-005 Malformed Canonical Response Handling

Purpose:

• Ensure malformed canonical files are surfaced explicitly.

Input:

• malformed JSON response file in canonical provider directory

Expected behavior:

• record malformed status
• do not silently include in primary comparison

Reason:

• Current audit depends on this behavior being stable.

RC-006 Machine Output / Human Output Consistency

Purpose:

• Prevent stale machine-readable output from disagreeing with current markdown reports.

Input:

• one run producing both raw-comparison-data.json and summary reports

Expected behavior:

• primary record count and data-quality counts agree across outputs

Reason:

• Current comparator state fails this consistency check.

RC-007 Structural Backbone Stability

Purpose:

• Ensure topology-preserving variants are not globally labeled unstable solely because of paraphrase differences.

Input:

• same graph topology expressed through baseline, renumbered, shuffled-edge-order, and identity-expanded variants

Expected behavior:

• backbone/topology result remains stable or mostly stable
• literal disagreement may remain visible

Reason:

• This is the core construct under test in EXP-001.

RC-008 Constraint Field Placement Sensitivity

Purpose:

• Measure whether semantically similar constraints moved across fields cause false disagreement.

Input:

• one response places "re-sorting follows differentiating input" under invariants
• another places it under validity conditions

Expected behavior:

• flag schema-placement sensitivity explicitly
• do not overstate it as full construct disagreement

Reason:

• This is the dominant current calibration problem in the constraint layer.

RC-009 Representation Metric Demotion Candidate

Purpose:

• Verify that free-text fields are treated as informational when style dominates signal.

Input:

• semantically similar summaries written at different compression levels

Expected behavior:

• AST presence remains comparable
• prose-heavy fields do not drive strong disagreement headlines

Reason:

• Representation calibration currently mixes compliance signal with stylistic noise.