I — Frame the Work · Chapter 03

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

Planning Under Uncertainty

What should a team promise when it still does not know which solution is appropriate?

The city has asked its engineers to investigate how the four-way intersection could serve road users better. The team can imagine several activities: revise the signal timing, add detectors, study physical movements, or explore adaptive control.

All of those may create work. Only some can responsibly be promised as product outcomes today.

A planning process that converts every idea into a scheduled implementation item would hide an important difference. Some activities are explorations, designed to discover whether an assumption holds. Others are commitments, whose outcome can be defined and evaluated with sufficient confidence.

We need to keep moving while making that distinction visible.

Begin with the next decision, not a backlog

The initial purpose is broader than any one solution: improve useful intersection operation without losing sight of competing road-user, physical and operational needs.

Before the team chooses the next investment, imagine a fictional concept review in which there is capacity for one substantial investigation. Three options are on the table.

Candidate investigation What is uncertain? Smallest useful result Dependency or stopping condition
A. Define movement conflicts Which intended movements and users must never receive incompatible indications? A reviewed conceptual movement/constraint map If the intended scope is unclear, clarify stakeholders and system boundary first
B. Compare arrival patterns Does the proposed fixed-phase abstraction respond differently to invented arrival clustering? A transparent model and reproducible trace Requires an agreed simplified function and input meaning; stop if the model cannot answer the question
C. Commit to new detection equipment Will hardware provide the information and behavior needed? An interface and feasibility hypothesis before procurement Premature until the intended sensing, command and installation needs are understood

Option C is not a bad idea. It is simply too early to call it a responsible product commitment in this hypothetical situation. Option A helps define what the system must consider. A bounded part of option B may then test a specific modeling hypothesis.

The proposed decision: clarify the conflict boundary and authorize a modest fixed-phase model investigation, while deferring detector procurement and adaptive-control commitments. This is a fictional planning decision, not a measured priority score or a municipal project approval.

Vision, initiative, increment and backlog are different things

Planning language helps only when each term carries a useful distinction.

A vision describes the intended longer-term outcome for users. A theme groups a direction of effort, such as improving confidence in traffic-control alternatives. An initiative identifies a problem worth addressing. A roadmap makes the intended progression and key dependencies visible. A delivery increment is a bounded capability or result that can be evaluated; a backlog holds the more detailed work needed to create it.

None of these labels makes uncertainty disappear.

The initiative might be “understand whether different fixed-time concepts warrant further study.” Its first increment could be a controlled, inspectable model comparison. That increment does not promise reduced real-world traffic delay.

A different initiative might concern physical installation feasibility. Its first result could be a reviewed interface assumption, not a delivered piece of equipment.

The question is whether each activity earns the commitment implied by its position on the roadmap.

Exploration and delivery can run together

It is tempting to picture a sequence in which all questions are resolved, then every requirement is specified, then the product is implemented. In an unfamiliar engineering problem, that is often neither practical nor desirable.

A team may investigate a new control concept while implementing a well-understood reporting improvement. It may build an emulator for a risky interface while another group prepares a bounded software change. Exploration and controlled delivery are concurrent loops, linked by evidence and decisions.

This is not a license for unrecorded improvisation. It is a reason to distinguish what is candidate, controlled, observed and approved at each point.

For our intersection, the fixed-phase model is an exploratory learning baseline. It can help the team ask a sharper question without becoming an approved traffic-control architecture.

Two roadmaps should influence each other

A product question may require a new engineering capability before it can be answered.

Consider the selected model investigation. The product-side question concerns how a control concept behaves with different arrivals. The engineering-environment work is to maintain a simple source definition, construct valid input scenarios, compute reproducible outputs and review what those outputs mean.

Those tasks deserve planning attention, but they are not products for their own sake.

The team should ask whether a new fixture, an automated check or a source relation will support one investigation or a broad set of recurring questions. A lightweight script may be enough for one study. A reusable integration test may be justified when similar changes occur frequently or have serious consequences.

“We build the machine that builds the machine” becomes a planning discipline when the team explicitly chooses which engineering capability is worth investing in now.

Plan for more than software

The software and modeling team may iterate on a candidate behavior quickly. New sensors or electrical adapters may need supplier interface agreements, physical samples and installation considerations. Operational evaluation depends on users, conditions and evidence that a simple model cannot supply.

A useful roadmap displays those different dependencies without pretending they share one cadence. It also reveals when an early model result can change a costly later commitment.

The team should not create fictitious durations to make the roadmap appear precise. It can still identify critical dependencies, which unknowns must be answered first and which commitments would be expensive to reverse.

Define a stopping and replanning rule

A decision to explore is more useful when everyone knows what would make the team change course.

For the fixed-phase comparison, the question is narrow: Can the same invented vehicle totals produce different end-of-window queues when arrival placement changes?

If the team cannot clearly define the arrival units, event order or model scope, it should stop interpreting results and repair the source definition before extending the experiment.

If the comparison is reproducible but does not address operational needs, the next action is not necessarily to build an adaptive controller. It may be to improve data, revisit stakeholder criteria or choose a different type of experiment.

A failed or inconclusive exploration can be progress when it reveals what is not yet fit for decision.

The plan should therefore track both completed work and knowledge gained, remaining uncertainties, decision owners and triggers for change.

Keep a small decision record

A short planning record for our fictional case could state:

Decision: Use a fixed-phase teaching model as an exploratory reference.
Reason: It makes one arrival-pattern question inspectable with a limited, understandable source set.
Not decided: Adaptive control, real signal timing, detector procurement, physical interface selection or operational readiness.
Needed engineering capability: Versioned model assumptions, scenario inputs and a transparent result trace.
Review trigger: A changed system boundary, invalid scenario data or findings that make the model unfit for the question.
Decision status: Provisional, for learning.

This is more useful than recording “simulation task complete” because it explains what the team has actually learned and what it has not authorized.

Why the information must become connected

Once several disciplines are investigating the intersection, assumptions appear in notes, models, interface drafts, diagrams and proposed checks. If two engineers use different versions of those definitions, even well-executed work can produce contradictory conclusions.

Our next problem is not how to add more backlog items. It is how to maintain engineering meaning and relationships so that the selected investigation and later changes can be understood.

Chapter 4, Engineering Knowledge as Code, begins with that need.

Engineering takeaway: Good planning under uncertainty distinguishes learning from commitment. It chooses the next important question, makes dependencies and stopping rules visible, and invests in engineering capabilities because they improve product decisions.


Related case: T1 source and defined information objects · Proposed model/data plan. The planning review is fictional and the model is educational.

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