I — Frame the Work · Chapter 02

All fifteen chapters
  1. The New Engineering Paradigm
  2. Introducing the Engineering Challenge
  3. Planning Under Uncertainty
  4. Engineering Knowledge as Code
  5. Learning Through Models and Prototypes
  6. Engineering Data: From Measurements to Decisions
  7. Spec-Driven Development Without Waterfall
  8. Directing AI Engineering Agents
  9. Engineering in Small, Reviewable Increments
  10. From Models to an Integrated Product
  11. Verification, Validation and Engineering Evidence
  12. Governance and Engineering Authority
  13. Continuously Integrated Engineering
  14. Managing Engineering Knowledge at Scale
  15. Choosing the Next Engineering Capability

Introducing the Engineering Challenge

One familiar system. Many engineering questions.

A city wants to improve how people move through a busy four-way intersection. Drivers report queues when vehicles arrive in bursts. People walking across the road want safe, understandable crossing opportunities. Maintenance staff want to know how to diagnose a fault when a controller and a signal indication disagree. Operations staff must keep the intersection usable during changes.

An adaptive controller sounds appealing. New detectors might help. Perhaps the existing signal sequence could be improved. But the city has not yet established which problem is most important, which movements must be protected or what evidence would justify a change.

This is the recurring fictional engineering challenge for the lab. It gives us a product question that is easy to picture but rich enough to connect software, electronics, physical systems, operational needs and engineering decisions. Nothing in the teaching case is an observed result from an actual intersection.

Start with outcomes, not a favored solution

The city’s broad ambition is to make the intersection serve road users better. That statement is not yet a testable requirement. Different stakeholders may mean different things by better.

Stakeholder Example concern Question that remains open
Drivers Long queues and unpredictable waiting Which traffic conditions and outcomes matter?
Pedestrians Safe and understandable crossing opportunities Which movements, conflicts and timing constraints must be protected?
Operators Predictable behavior and manageable changes What conditions and procedures define acceptable operation?
Maintainers Fault diagnosis and equipment state How will commanded, acknowledged and observed states be distinguished?
Engineering team A defensible concept and credible evidence Which experiments can answer which product questions?

These concerns cannot be reduced to one queue metric. Optimizing an abstract model without defining crossing safety, interfaces and intended operation could make the wrong problem easier to solve.

The first decision is therefore not which controller to purchase. It is what the team must understand before it can make that decision.

Establish a whole-system boundary

The conceptual product includes the control function, signal indications, relevant electrical and communication interfaces, physical installation, road users, maintenance, operating procedures and environmental conditions. Software is an important part, but not the complete system.

Our earliest executable model is deliberately smaller. It considers a simplified fixed-phase signal sequence and invented vehicle arrivals. It can help us reason about some queue behavior under stated rules. It cannot establish safe crossing times, real traffic performance, controller reliability, hardware timing or compliance with traffic regulations.

Keeping the product boundary and the model boundary separate is essential. The model may answer a narrow engineering question without representing the product well enough to approve a real-world change.

The first learning question

Imagine two scenarios with the same total number of arriving vehicles but different arrival patterns. Would a deliberately simplified fixed-phase model produce different queues at the end of the observation window?

That is a suitable initial learning question, not a promise of improved traffic throughput. The team needs a defined phase sequence, a meaning for arrivals and departures, explicit event-order rules and traceable results. The answer may show something useful about the abstraction—or expose missing assumptions.

We will not solve that question here. Later chapters will introduce the controlled engineering information, executable model and data needed to investigate it properly.

Two kinds of work follow from the same question

On the product side, the team must define relevant road-user movements, intended behavior, interfaces and acceptable trade-offs. It must distinguish what is known from what is merely proposed.

On the engineering side, the team needs ways to manage shared definitions, run bounded experiments, inspect outputs, test changes and retain decisions. A simple script may be sufficient for one experiment; a reusable test environment may be worth developing when the same checks recur.

This is where the two-system idea from Chapter 1 becomes practical. The engineering environment should grow in response to an actual product need, and the improved environment should make subsequent product decisions more reliable.

How the example develops through the course

We will use this same intersection from start to finish, with each chapter adding a distinct engineering concern:

  • Planning and information: choose the next uncertainty to investigate, then preserve definitions and relationships so disciplines share the same meaning.
  • Models, data and specifications: make a bounded behavior executable, inspect what it reveals, and define a controlled next increment without pretending to know every answer.
  • Agents and integration: delegate suitable work with feedback and review, connect software to electrical and physical interfaces, and discover discrepancies that a pure software model might miss.
  • Evidence, authority and evolution: establish what has been verified and validated, determine what may responsibly be released, and decide which engineering capability is worth improving next.

The storyline is educational rather than a prescribed project sequence. Real teams must consider integration, verification, safety, security and operational needs from the beginning, even though later chapters examine those topics in greater depth.

What the example will—and will not—prove

The lab uses synthetic scenarios and a limited teaching model to make engineering reasoning inspectable. A model output or automated check can support a claim about that model with those inputs. It does not demonstrate that an actual intersection controller has been built, connected to signals or qualified for operation.

As the story develops, we will distinguish proposed changes from executed checks, software behavior from physical behavior, and model evidence from operational acceptance. These boundaries are part of the engineering lesson, not obstacles to learning.

The next decision

The city has multiple plausible improvements, but insufficient grounds to choose one. The team needs to decide which uncertainty matters most, what investigation would reduce it, and what it can responsibly commit to next.

Chapter 3, Planning Under Uncertainty, starts there. It asks how to plan meaningful progress without disguising an exploration as a guaranteed product result.

Engineering takeaway: A shared example is useful only when its stakeholders, system boundary and evidence limits remain visible. The four-way intersection gives us one coherent product story—not a preselected solution—to carry through the rest of executable engineering.


Reference case: Fictional intersection scope · T1 educational sources. The model and any AI-agent workflows in later chapters are bounded teaching examples, not a physical traffic-control validation.

Published after initial author review. This fictional teaching case is not physical validation. Interactive V3 tools remain in development. Chapter source