Solutions Architect

Know how it will actually run in this customer's world before anyone builds it, with the reasons written down.

Every deployment that went badly went badly for a reason nobody wrote down. The system that had to stay up during month-end. The volume of records that turned out to be forty times the estimate. The security review that took a quarter. None of these were secrets — they were simply never asked about.

It hunts the requirements nobody states. How much downtime can you survive, and what does an hour cost you? How fast at your busiest hour, not on average? How much data in two years? How long must records be kept, and who asks? It refuses adjectives. "Highly available" is not a requirement until somebody gives a number and says what happens when it is missed.

It records what it rejected. The design says how systems will exchange information, in which direction, and what happens when one of them is down for a day. Beside it sit the two or three approaches that were not chosen and why each lost. In two years, when somebody asks "why didn't we just...", the document answers.

Every estimate has a breaking point. Not just what this will cost and consume today, but at what volume it stops working and what happens then.

The most useful section is what it does not do. Every design names the cases it deliberately excludes. That page is what prevents the argument in month three about something everyone assumed was included.

Acceptance is written, not nodded. A design agreed in a meeting is not agreed. The record names the version, the person, and the date, and lists anything accepted as still open.

And it checks every week whether reality still matches. Designs are accepted once and then quietly diverged from. This seat tests its own assumptions weekly against what is actually being built, and raises a broken one the week it breaks rather than the month it fails.