dekc_grade.py / dekc_doctor.py cannot grade a walk inside a mixed SAC+PKC+DEKC brain #39

Summary

Grade and doctor iterate every markdown concept in the bundle. A shared second brain with SAC reverse-engineering (~4k Module, ~3k Function, ~2k Class) plus PKC tickets cannot produce a Fabric-walk RE score. Doctor reports concepts: 14988 and tables→BO: 0.07 dominated by pre-existing NDQR PDFs, not the walk.

dekc_grade.py also crashes (dict descriptions) before scores. Producer self-grade had to be hand-written.

The RE rubric is defined per walk. There is no --focus, --since, --tag dekc, or --walk-id to score only this turn’s nodes.

Expected

• --prefix knowledge/semantic,knowledge/tables/gold- or --tag fabric-walk-2026-08-24.
• Ignore types this plugin does not own (Module, Function, TicketLink, DecisionRecord) unless --all.
• Doctor: separate “DEKC technical” vs “bundle-wide OKF.”

Suggested fix

Filter list_concepts by TYPE_TO_DIR keys this pack owns (29 nouns) by default. Add --since TIMESTAMP using frontmatter timestamp / WriteEvent.

Found on: DEKC 0.4.1 (GitHub Latest 2026-08-24), mixed PKC+SAC+DEKC second brain, Azure Fabric reverse-engineering walk.
Not a sample-knowledge issue. Repro used live capture scripts from spillwave-second-brain cache data-engineering-knowledge-capture/0.4.1/.