From the prologue
The dirty price
Two days on one number
In February 2023, at the sticky end of a Sydney summer, the other engineer and I spent two days trying to find out why our bond prices were wrong.
They were only slightly wrong, which made it harder. We had built a pricing screen, the first piece of our software that looked like something a client might one day pay for, and for almost every bond on it our price came out a little higher than the price Rhys had written in his notebook. Sometimes the gap was a few cents per 100 dollars of face value. Sometimes it was more than a dollar. It was never the same two days in a row.
We worked from a couple of desks in a shared office in the city. The startup was building software to price, distribute and trade bonds electronically for institutional investors in Australia and Asia. I had joined a few weeks earlier as chief technology officer, a title that in a company of five meant writing a lot of the code and descaling the kettle. The building’s air conditioning went off at six. By seven the humidity had found its way back in, and we were still there.
We did what engineers do with a bug. We assumed it was ours and went looking in the usual places. Floating-point rounding came first, because it is always the first suspect and hardly ever guilty.