research-document

FE-012C Semantic Repeatability Report

FE-012C Semantic Repeatability Report

Purpose

Add a layered semantic grammar comparison beside the existing literal comparator.

This report does not replace syntactic comparison.

Literal Drift Summary

  • Stable: 1/45
  • Elaboration drift: 8/45
  • Structural drift: 36/45
  • Missing or invalid: 0/45

Semantic Drift Summary

  • Stable: 6/45
  • Elaboration drift: 0/45
  • Structural drift: 38/45
  • Unclear: 1/45

Cases Where Literal Drift Disappears Under Semantic Comparison

Packet ID Provider Literal Semantic Rules Triggered Notes
FE-012C-P010 gpt elaboration_drift stable none Semantically stable. Rules triggered: none.
FE-012C-P010 claude elaboration_drift stable none Semantically stable. Rules triggered: none.
FE-012C-P011 claude elaboration_drift unclear none Semantic normalization reduces some drift, but ambiguity or candidate missing primitives still warrant human review.
FE-012C-P007 gemini elaboration_drift stable none Semantically stable. Rules triggered: none.
FE-012C-P010 gemini elaboration_drift stable none Semantically stable. Rules triggered: none.
FE-012C-P013 gemini elaboration_drift stable none Semantically stable. Rules triggered: none.

Cases Where Semantic Drift Remains Real

Packet ID Provider Literal Semantic Notes
FE-012C-P001 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P002 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P003 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P004 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P005 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P006 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P007 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P008 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P009 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P011 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P012 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P013 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P014 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P015 gpt structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P001 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P002 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P003 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P004 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P005 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.
FE-012C-P006 claude structural_drift structural_drift Backbone or control-flow structure still differs after semantic normalization.

Uncertain Cases Needing Human Review

Packet ID Provider Literal Notes
FE-012C-P011 claude elaboration_drift Semantic normalization reduces some drift, but ambiguity or candidate missing primitives still warrant human review.

Interpretation

  • Semantic comparison is provisional tooling, not a proven equivalence theory.
  • Literal differences remain part of the record even when semantic drift decreases.
  • Backbone agreement, elaboration drift, and control-flow agreement should be interpreted together rather than in isolation.