Models are often built as black boxes. A user can explore the dynamics of the model, trying different actions and viewing the results. But the internals are hidden: the user cannot see how his actions produce the results.

A black box model sets up two very different roles: the modeler and the user. The modeler knows how the model works. The user is to be content with what the model does. Pay no attention to the man behind the curtain.

Users distrust all models, but particularly black box models. They don’t believe the results until they see how the results were made. How does the logic work? What assumptions does the model depend on? How sensitive are the results to those assumptions? What happens if other assumptions were made?

It’s good to make the logic visible, to turn the black box into a white box. A user can then trace the logic, to see how his results were formed from his actions. He can learn from the model.

But users are not modelers. It’s not enough for the model logic to be visible. For an everyday user to understand a model, it’s better if the the model is made comprehensible to everyday people. Complex definitions should be refactored into simple components that can be understood by anyone. English-language explanations should be written.

Models have assumptions, often hundreds of assumptions. Some assumptions reflect solid data. Some reflect the opinions of many experts. Some assumptions are merely educated guesses. It’s even better if all these assumptions are made visible to the user, not just the numeric values assumed but who made that assumption, and why.

Can the assumptions be changed? Can a user alter an assumption when new data arrives? Can he change an assumption because he believes the original value is mistaken and he knows better? Can he change an assumption just to test what happens, to see how sensitive the results are a particular assumption? It’s best if all the assumptions are modifiable by users. It’s best if the model has a user interface for the assumptions, allowing any of them to be changed.

Visible logic, comprehensible structure, modifiable assumptions: we are now a long way from the black box. Open-book models are those models that have all these properties: that are visible and comprehensible to everyday users, and that have assumptions that can be examined and changed by those same users.  The term “open-book model” is intended to suggest the analogy to open-book management. Just as the practitioners of open-book management enlist everyday corporate employees by revealing the secrets of the company’s finances, open-book models enlist everyday users of a model, to win over their trust.

Open-book modeling breaks down the barriers between modelers and model users. Users can see how everything works. Pay attention to the man behind the curtain.

Open-book modeling is difficult. It is much easier to create a model that is opaque, then one that everyday users can explore and change. But the result is worth the extra effort. Open-book models are far more valuable than black box models.