dekc_walk.py only walks a SQL/parquet directory; Fabric REST / TDS profile is documentation-only #38

Summary

The reverse-engineering orchestrator and design doc describe an Azure Fabric cloud profile (lakehouse, pipelines, semantic model, Power BI). The script still walks a local directory of *.sql / *.dax / parquet.

On NDQR:

• dekc_walk.py ../ndqr/sql/fabric --dry-run found 18 SQL files (ndqr_wh DDL only) and skipped them. It cannot see OneLake, 705 lakehouse tables, 20 SemanticModels, or 141 pipelines.
• Live inventory had to be done with az + Fabric REST + Power BI REST outside DEKC, then hand-fed to dekc_capture.py.
• CREATE TABLE DDL produces no from-sql edges (dekc_lineage.py only emits when a write target and FROM both exist).

Expected

A Fabric profile should at least:

1. Ingest a workspace items JSON (Report, SemanticModel, Lakehouse, Warehouse, DataPipeline, Notebook).
2. Ingest Power BI report→dataset bindings and datasources (database grain).
3. Optionally ingest an INFORMATION_SCHEMA / live inventory JSON without inventing COUNT metrics.
4. Map Fabric Report vs DEKC Dashboard/Report; default SQL-endpoint SemanticModels vs curated models.

Until then, skills should say: dekc_walk.py is a filesystem mirror walker, not a Fabric control-plane scanner.

Code

scripts/dekc_walk.py walk_lake() — rglob("*.sql"), layer folder hints.

agents/reverse-engineering-orchestrator.md — cloud profile fabric: Eventstream, Pipelines, Lakehouse, semantic model + Power BI. No script implements that list.

Suggested fix

• dekc_walk.py --fabric-items path.json / --pbi-bindings path.json.
• Keep filesystem walk for git SQL mirrors.
• from-sql: do not claim “no lineage” for CREATE TABLE; report “DDL-only, no FROM.”

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/.