Twenty-five years of Windows applications that had to work

The short version

From 2006 to 2023 the job was large enterprise WPF: trading and risk applications for investment banks, where a slow grid is not a cosmetic problem and a threading bug is not a ticket for next sprint. That means seventeen years of the specific things that go wrong in serious desktop software — binding cost, dispatcher contention, virtualisation, allocation pressure, custom controls that have to render thousands of updates a second without dropping a frame.

Along the way: Goldman Sachs, UBS, BNP Paribas, Barclays Capital, Morgan Stanley, Citi, BNY Mellon, Merrill Lynch and Crédit Agricole CIB, plus an eight-year stint at BP on Gordon, an oil-trading decision-support system. More recently the work has broadened into .NET APIs, React and TypeScript front ends, AWS and Kubernetes — useful, because most desktop applications now sit on top of a service estate that has its own problems.

Number Six exists because that combination is getting rare. Fewer people are learning WPF and WinForms every year, while the amount of business-critical software written in them has not gone anywhere. Somebody has to keep it working.

Why “Number Six”

Six is the number of the diagnostic: the sixth question, the one that gets asked after the obvious five have not explained the behaviour. It is also, conveniently, a shape that looks like a line curling round into a closed, stable form — which is a fair description of what we do to a codebase.