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Historical Status: Superseded as current-state guidance by Comparator v3.1 implementation and regression results.

Primitive Calibration Audit

Scope

This audit reviews primitive disagreement behavior reported in:

• comparison/generated/primitive-stability-report.md
• canonical EXP-001 response files

Current output:

• primitive_sequence: disagreement
• transitions: disagreement
• dominant_primitive: disagreement
• candidate_missing_primitives: disagreement

Assessment

The primitive comparator is currently too strict for the way models express procedural structure.

The reported disagreement is mostly not evidence of true procedural divergence.

Source Of Disagreement By Primitive Field

Primitive sequence

Primary causes:

• wording
• abstraction level
• ordering

Examples:

• node-centric sequences
  • START -> N1 -> N2A/N2B/N2C -> N3 -> N4
• action-centric sequences
  • expand -> converge -> differentiate -> re-sort -> terminate or loop
• mixed sequences
  • node names plus action labels in one list

Audit classification:

• mostly wording and abstraction
• occasional ordering
• rarely actual procedural difference

Transitions

Primary causes:

• representation form
• ordering
• compression versus expansion

Examples:

• expanded object form
  • one transition per edge
• compressed string form
  • N1->N2A|N2B|N2C
• pass-through expansion in P001D
  • some models include N1a/N1b/N1c
  • others compress directly to N2A/N2B/N2C

Audit classification:

• mostly representation and ordering
• some actual procedural difference in pass-through handling

Dominant primitive

Primary causes:

• wording
• ontology

Examples:

• "conditional branching followed by convergence and conditional iteration"
• "expand into candidate paths, merge, differentiate, re-sort, and either continue looping or terminate"
• "branch-converge-reassess"

Interpretation:

• These often point to the same procedural role cluster.
• Current disagreement here is mostly a normalization problem, not a construct problem.

Candidate missing primitives

Primary causes:

• missing-primitive speculation style
• comparator strictness
• ontology gaps

Examples:

• one model lists threshold definition
• another lists path-selection rule
• another lists evaluation metric

Interpretation:

• This field behaves more like a reviewer note than a stable comparable measurement.
• It should not currently carry the same calibration weight as explicit primitive sequence or transition structure.

Cause Matrix

Conclusion

Primitive disagreement is currently exaggerated by normalization and representation sensitivity.

Scientifically useful:

• pass-through expansion versus compression in P001D
• explicit versus implicit loop transitions

Not yet scientifically stable:

• dominant primitive
• candidate missing primitives

Calibration implication:

• primitive sequence and transitions may remain score-bearing after calibration
• dominant primitive and candidate missing primitives should be treated as lower-confidence until comparator normalization improves