Well-Defined vs. Ill-Defined: Why Infrastructure Decisions Resist More Data

Two Kinds of Problems

In 1973, Herbert Simon published a distinction that has quietly shaped decision science for five decades. Well-structured problems have a defined initial state, a defined goal state, and a set of legal moves connecting them. Solving a system of equations. Routing a shipment. Optimizing a schedule. More information about the variables makes the answer more precise, and precision converges toward a single correct solution.[1]

Ill-structured problems do not have this shape. The initial state, the goal state, and the path between them are all incompletely specified. Simon's own framing was that the boundary between the two categories is "vague, fluid, and not susceptible to formalization," but the distinction still holds where it matters most: whether the problem-solving process has a defined target to converge on.[1][2]

Most infrastructure capital decisions are ill-structured problems managed as though they were well-structured ones.

Why the Loop Happens

Decision analysis has a formal answer for why this matters, and it predates Simon by seven years. In 1966, Ronald Howard at Stanford developed information value theory: information only has value relative to the decision it informs. The value of eliminating an uncertainty is a function of how that uncertainty affects an outcome you have already defined.[3][4]

This has a direct and uncomfortable implication. If the outcome is not defined, information does not have a computable value. It cannot converge you on an answer, because there is no answer to converge on yet, only more description of a space that still has no target.

This is the mechanical explanation for the loop infrastructure operators find themselves in. Acquire more data. The data does not resolve which scenario matters, because "which scenario matters" was never defined against an outcome. Acquire more data again. The problem is not the volume of information. The problem is that value of information theory has no defined objective function to operate against.

What This Looks Like in Infrastructure

The consequence shows up as capital committed on assumptions that were never tested against a defined outcome, and it is visible at scale. In the first quarter of 2026 alone, at least 75 U.S. data center projects worth approximately $130 billion were blocked or delayed, roughly matching the entirety of 2025's total disruption in three months.[5] The number of active local opposition groups tracking these projects more than doubled over the same period, from 396 to 833 across 49 states.[5]

These are not primarily data shortages. Every one of these projects had access to engineering models, market forecasts, and site data. What they lacked was a decision process that started by defining the outcome being protected, the scenarios that could threaten it, and which uncertainties were actually decision-relevant before capital was committed.

The Practical Implication

Simon's framework and Howard's framework point to the same operational conclusion from different directions. Simon shows that ill-structured problems require constructing the problem representation before a solution process can even begin. Howard shows that information has no computable value until an outcome and a decision are defined. Neither one is solved by aggregation. Both are solved by sequencing: define the outcome first, then determine which data is actually worth acquiring against it.

The future of infrastructure decision-making will not be determined by who aggregates the most data. It will be determined by who defines the decision first.

---

References

[1] Simon, H. A. (1973). The Structure of Ill Structured Problems. Artificial Intelligence, 4(3-4), 181-201.

[2] Reed, S. K. (2016). The Structure of Ill-Structured (and Well-Structured) Problems Revisited. Educational Psychology Review, 28(4), 691-716.

[3] Howard, R. A. (1966). Information Value Theory. IEEE Transactions on Systems Science and Cybernetics, 2(1), 22-26.

[4] Raiffa, H., & Schlaifer, R. (1961). Applied Statistical Decision Theory. Harvard Business School.

[5] Data Center Watch (10a Labs), Q1 2026 Report.