DUKE FIELD GUIDE · CONTEXT ENGINEERING
THE ENTERPRISE
CONTEXT LAYER
SPECIFICATION
What an AI data agent should know before it writes SQL.
A practical specification for the business rules, metric definitions, schema knowledge, join paths, exceptions and verified examples that prevent an AI data agent from guessing how your company interprets its data.
8
SECTIONS
6
PAGES
1
REUSABLE TEMPLATE
Built for AI data agents, Text-to-SQL systems and enterprise analytics teams.
SPECIFICATION PDF · 6 pages · Work email required
WHAT GOES INSIDE
01
BUSINESS MEANING
- Global rules
- Business vocabulary
- Metric definitions
The decisions that cannot safely be inferred from the database.
02
DATA STRUCTURE
- Relevant schema
- Representative values
- Canonical join paths
Enough structure for the agent to navigate the model without exploring blindly.
03
OPERATIONAL REALITY
- Exclusions
- Temporal rules
- Known traps
- Conventions
The details that make technically valid queries produce wrong business answers.
04
VERIFIED EVIDENCE
- Known-good queries
- Critical reminders
- Evaluation cases
Examples and tests that make the context executable rather than merely descriptive.
THE EIGHT-PART FILE
- 01GLOBAL RULES
- 02BUSINESS LANGUAGE
- 03METRICS & FORMULAS
- 04DATA MODEL
- 05JOIN PATHS
- 06TRAPS & CONVENTIONS
- 07VERIFIED QUERIES
- 08CRITICAL REMINDERS
A context layer is not everything the company knows. It is the information the AI must not be allowed to guess.
TREAT IT LIKE CODE
IF CONTEXT CAN CHANGE THE ANSWER,
CONTEXT IS PRODUCTION LOGIC.
A stale context file rarely throws an error. It produces a plausible answer using an outdated interpretation.
DOWNLOAD
THE ENTERPRISE CONTEXT LAYER SPECIFICATION
Want to see how this works on your own data model?
DUKE builds the governed layer between enterprise data, business meaning and AI analysis.