Fix or rewrite

The team says rewrite. The contractor says fix.

Both sides have grounds, and both sides have an interest. Rewriting is more interesting and more legible to an engineer. Fixing is cheaper this quarter and more legible to whoever pays. The argument runs in a language with no shared unit of measure, and so it does not end.

Both doors are counted in the same two currencies. Technical risk — what exactly can break and what stops working for people when it does. FTE — how many person-months each door costs and when those people come free.

It is counted part by part, not as a whole system. The answer is almost never a single one: usually part of the system is cheaper to remove and part is more expensive to touch at all. The boundary between them is the real result.

A third door is considered. Besides fixing and rewriting there is almost always a third: freeze it as it is and build alongside. It rarely comes up in the argument, because neither side likes it.

The price of waiting is named. A deferred decision is a decision too, and it has its own price in the same two currencies.

Two or three doors, counted the same way, each with its price and its risk. You choose: this is a decision about money and deadlines, not about technology, and it belongs to whoever answers for them.

If I have a recommendation, it is given on its own line and marked as a recommendation, not as the output of the calculation.

Eight months of someone else's work went onto the table in two currencies — risk and people. Finishing it came out more expensive than removing it: eighty percent of the code went, the rest was completed, and the first machine reached production two weeks later.

The same arithmetic gave the opposite answer at a bank. Business logic living in the database was moved in small steps and not rewritten in full: a target architecture is not by itself a reason to rewrite. That was worth saying out loud, because "but we have agreed on a different approach" sounds convincing and costs a great deal.

A verdict. I do not decide which door you open — I make sure both are counted the same way and that you have a choice with a known price.

And there is no promise that rewriting will turn out cheaper. More often it is dearer than the team thinks and dearer than it sounds in the deck.

Name the part of the system the argument has run longest about, and one number that can be measured: release time, incident count, time to ship a new feature. We count both doors from there.