databricks-SL · Capstone · Semantic Layering (benchmarked vs DeepMind reasoning principles)

Semantic Layering

The capstone. Turn gold tables into meaning an agent can reason over: an entity model, named relationships, an executable semantic graph, governed metrics, a vocabulary, a governance matrix, a retrieval corpus, MCP tools, and an agent eval set. Fourteen gates. Definitions of a semantic layer pass none of them.

Not submitted

Your mentor

Dr. Lena Sørensen

active

This is the lesson that matters most. A dashboard layer is not a semantic layer. An agent has to know what an entity is, how entities relate, what a metric really means, and what it's allowed to see. Build that — I'll check it against all fourteen gates, not against your phrasing.

The deliverable · 14 gates

▸Semantic layer definition (three pillars)

▸Entity catalog with entity_class

▸Table-to-entity map (row grain + key)

▸Column-to-property map (semantic roles)

▸Named relationship graph (subject-verb-object + cardinality)

▸Executable semantic graph (valid + invalid paths)

▸Governed metric catalog

▸Vocabulary (synonyms → objects)

▸Governance & sensitivity matrix

▸Semantic retrieval corpus design

▸MCP / semantic API tool spec

▸Agent evaluation set (expected reasoning paths)

▸Short design explanation

Your work · artifact bundle

0 files

Empty workspace. Add files, upload a folder, or load a demo bundle to start.

Submit the bundle to the kernel. It grades the artifact gate-by-gate — the mentor never gives a verdict.

Kernel evaluation

PASS = ∏ γᵢ over 14 gates · any failed factor ⇒ FAIL

Recognition Filter

Layer Definition

Entity Model

Table to Entity Map

Column to Property Map

Named Relationship

Graph Reasoning

Metric Grounding

Context & Vocabulary

Governance & Sensitivity

Vector Retrieval

MCP / Semantic API

Agent Evaluation

Evidence Completeness

World output

Semantic Reasoning Layer

Locked until every gate passes. Nothing downstream can attach to an unverified layer.