Fil Salustri's Design Site

Site Tools


Eliminate the Hay

If you don't know how to find the best choice of many, try eliminating as many of the bad choices as possible.

Sometimes, searching for an appropriate solution among many possibilities cannot be automated. In conventional searches, you can tell from any given item which way one should move through the solution space to find a better solution, thus inevitably finding the “best” solution.

Other times, however, you don't know how to get from one solution to a better solution, knowing that a “best” solution can be found. These cases are like looking for a needle in a haystack: looking in any one spot in the haystack does not tell you anything about where else to look, even though you would know the needle once one has found it.

Design problems are like the needle in a haystack problem. In all but the most trivial situations, it is not possible to tell a-priori whether a given design is “best” for the situation, or what other design to consider as better than the current one. This is especially true during early design, when so very little is known about any potential design concept. This means that conventional search algorithms do not work in design.


Instead of looking for the one “best” design in a possibly huge set of possible designs, look for and eliminate the worst designs.

This is analogous to looking for the hay in the haystack, rather than the needle. If you can eliminate most of the hay, it will be much easier to find the needle. It's easier to find bad designs (hay) because there's more of them, and every time you make the set of possible designs smaller (by eliminating bad designs), you shorten the time needed to find the “best” design.

It's also easier to identify a bad design than it is to identify the “best” design. The “best” design must be well-balanced; that is, all its parameters and behaviours must be nearly perfectly tuned to its situation. On the other hand, a bad design is any design that has one suboptimal parameter or behaviour, or one violated constraint.

To make this work, you must be able to identify a parameter, behaviour, or feature the value of which you can easily determine. Every time you find such an attribute, you're essentially eliminating every design with that attribute.

Specific techniques to do this include: mismatched ideas.


When attempting to reduce the number of design concepts one has (e.g., by eliminating inconsistent embodiments, the concepts are still quite vaguely defined. It is possible that a design that is culled with this technique was actually a good concept. It is important, therefore, to carefully justify and document the criterion by which you identify bad designs. If it turns out that the criterion was poorly defined, you can go back, adjust the criterion and repeat the operation. Doing so will of course incur a resource cost that may adversely affect lead time and development costs.

design/eliminate_the_hay.txt · Last modified: 2020.03.12 13:30 (external edit)