Comparison Evidence

This page complements Comparisons with the generated evidence files that back the reviewer-grade matrices.

Primary method citations and official upstream package links are collected in Method References.

Generated evidence files

  • docs/assets/generated/evidence/comparison_evidence.json
  • docs/assets/generated/evidence/comparison_capability_matrix.csv
  • docs/assets/generated/evidence/comparison_install_matrix.csv
  • docs/assets/generated/evidence/comparison_family_fairness.csv
  • docs/assets/generated/evidence/comparison_agent_matrix.csv
  • docs/assets/generated/evidence/workflow_comparison.json

Regenerate them with:

python benchmarks/software_comparison/generate_comparison_evidence.py
python examples/workflow_comparisons/compare_specialist_glue_vs_detime.py

Capability emphasis

The matrices are designed to support one narrow claim:

De-Time is not the deepest specialist package in every family; it is the most workflow- and machine-contract-oriented decomposition layer among the compared packages.

This is why the generated comparison tables prioritize:

  • unified config and result objects
  • machine-readable catalog fields
  • compact serialization modes
  • CLI and profiling support
  • MCP / tool surface
  • explicit maturity labeling

Workflow-level comparison

The workflow comparison artifact contrasts two paths:

  • a multi-package specialist glue workflow that mixes multiple result objects and package-specific APIs
  • a De-Time workflow that keeps the result contract, routing surface, and serialization story under one import path

This is the software abstraction argument behind the package, not a benchmark leaderboard claim.