research-document

Comparator v3.1 Regression Report

Comparator v3.1 Regression Report

  • total tests: 33
  • passed tests: 33
  • failed tests: 0
  • warnings: 0
  • blocking failures: 0
  • mode: full

Exact Failed Test Names

  • None

Raw Response Hash Verification

  • verified: yes
  • files checked: 13

Regression Checks

ID Severity Name Status Evidence
RG-001 critical broad-family resemblance with explicit negation becomes partial PASS Recognition classifier regression
RG-002 major unrecognized maps to unknown PASS Recognition boundary regression
RG-003 major empty recognition remains not_recognized PASS Recognition empty-field regression
RG-004 critical raw responses are unchanged PASS 13 response hashes compared before/after run
RG-005 critical all generated outputs share one immutable run state PASS Run state ID is written from one in-memory result set
RG-006 major structural backbone stability is separated from wording variance PASS Backbone=mixed; Literal=mostly_stable
RG-007 major neutral identity-node expansion is handled as elaboration PASS Backbone profile mixed
RG-008 major primitive reporting separates role/transition stability from style variance PASS Primitive roles, transition backbone, and style variance are all reported
RG-009 major constraints are compared conceptually across field placement PASS Conceptual=disagreement; placement=high
RG-010 major free-form representation prose is informational rather than score-bearing PASS Representation prose metrics demoted to informational status
RG-011 major tolerant parsing is explicit and auditable PASS Tolerant parse events appear in data-quality output
RG-012 critical EXP-001 outputs regenerated under generated-v3.1 PASS /Users/kevinmiller/dev/Framework-engineering/research/evidence-runs/ECR-000003-representation-sensitivity/experiments/EXP-001-topology-perturbation/comparison/generated-v3.1
RG-013 critical no output consistency mismatch PASS primary=12; machine=12; dq=5

Promotion Gate

  • all critical tests pass: yes
  • all major tests pass: yes
  • no output consistency mismatch: yes
  • no false-recognition regression: yes
  • no raw-response mutation: yes