Module 20 Cheat Sheet

AI SHOULD
=========

extract
organize
compare
implement
test
review
trace
detect contradictions


AI MUST NOT
===========

invent business semantics


SOURCE OF TRUTH
===============

repo artifacts:

business-rules.md
mapping.csv
OpenAPI
architecture.md
test-matrix.md


REQUIREMENT STATES
==================

CONFIRMED
ASSUMED
OPEN
DEPRECATED


TRACEABILITY
============

BR-001
↓
mapping
↓
code
↓
test


AI SKILLS
=========

requirements analyst
mapping reviewer
architecture reviewer
DataWeave reviewer
Salesforce reviewer
resilience reviewer
test designer
regression analyzer
security reviewer
observability reviewer
release reviewer


CORE AI RULES
=============

no guessing

external IDs / identity explicit

null semantics explicit

retry ⇒ idempotency review

timeout = ambiguous

no assumed rollback

avoid N+1

verify real SF separately

no secret logging

flag contradictions

The interview-worthy sentence tying both modules together is:

“I like to make integration requirements executable and traceable: business rules get IDs, mappings reference them, acceptance scenarios are generated from them, and automated tests trace back to the same rules. AI can automate much of the requirements analysis, mapping review, implementation and regression-impact analysis, but I explicitly prohibit it from inventing missing business semantics—it must surface those as Product decisions.”

And one thing I would add to the two-day Mule exercise: have the AI create the mapping matrix and business-rule/test matrix before it writes the first DataWeave script. That forces the implementation to be requirements-driven instead of letting generated code become the de facto specification.