| name | dekc-grade |
| description | Grade reverse-engineering quality with automated rubric scores and adversarial judge protocol. Use after walks, before claiming RE complete. |
DEKC grade (adversarial RE quality)
Automated baseline
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_grade.py" --repo . --bundle knowledge
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_grade.py" --repo . --bundle knowledge --json
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_grade.py" --repo . --bundle knowledge --write
Default scores DEKC nouns only (29 types). Mixed SAC+PKC brains (~15k Module/Function nodes) will not dominate.
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_grade.py" --repo . --bundle knowledge \
--prefix semantic,tables/gold-,reports,dashboards,ingestion
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_grade.py" --repo . --bundle knowledge --tag fabric-walk-2026-08-24
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_doctor.py" --repo . --bundle knowledge
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dekc_doctor.py" --repo . --bundle knowledge --all
Dict-valued description fields (bad YAML / mixed bundles) are coerced; grade does not traceback.
Adversarial loop (required for acceptance)
- Spawn lineage-skeptic, business-skeptic, stream-job-skeptic, coverage-skeptic
- Aggregate with re-adversary-judge against
evaluation/reverse-engineering-rubric.md (threshold 0.75)
- On fail → orchestrator revises (capture evidence or retract claims) → re-grade
- On pass →
dekc_index.py build
Orchestrators: data-lake-walker, reverse-engineering-orchestrator.
Rubrics
evaluation/reverse-engineering-rubric.md
evaluation/lineage-integrity-rubric.md
evaluation/business-fidelity-rubric.md
evaluation/stream-job-landing-rubric.md