Decision Under Uncertainty

Architects are asked to commit before the information exists — load is unknown because the product has not launched, requirements will change, the technology is unfamiliar. Waiting for certainty is itself a decision, and usually a worse one.

The moves that work better than deliberation. Ask what would have to be true for each option to be right, then test the cheapest of those assumptions directly rather than debating them — a half-day spike frequently settles an argument that would otherwise run for two weeks.

Choose by the cost of being wrong rather than by expected value. Where one option fails recoverably and another fails catastrophically, the asymmetry dominates the probabilities, and this is the reasoning that should be written down.

Defer what can be deferred without cost. Not every decision needs making now, and identifying the last responsible moment — the point beyond which deferral eliminates an option — is a distinct skill from deciding.

Design for the decision being wrong: an abstraction, a flag, a facade, a pilot. Optionality has a price and it is usually less than the cost of the wrong permanent choice.

And write down what would change your mind. That converts a decision into something reviewable against evidence rather than a position to defend, which is the difference between an architecture that learns and one that accumulates commitments.