Anticlue

The 1 - 2% margin of error

Joel Spolsky has written an excellent article on software craftsmanship.

The article elaborates on why fixing the last 1% of bugs is enormously expensive.

What are the last 1% or for that matter 2% of bugs? That depends on your focus, if you work for a place that is developing a product for mass release, these issues are normally caught in testing the unique cases, in joel's example ftping a 128MB content update file. If you work for an organization that's focus is something else than releasing software, these are the issues that your users bring to you or you find out from some other mechanism. They were not identified in testing. Why? well first, in house developers don't have access to all of the different browsers and operating systems to test things on. If we are lucky we have 2 - 3 setups. If we are really fortunate one of those is a mac. Secondly, we don't test for every possible outcome. It takes too much time, and we never ever capture all of them.

So here we are with in-house developed software, that works at 388 workstations, but not at 8. What do you do? Do you take the time to fix it for the 8 workstations, even though that may take anywhere between hours to weeks. If the fix time is in day multiples, you will need to reschedule all of the projects in the pipeline to resolve this issue. (Resources are limited) If you choose not fix it, you will leave the users with an embittered experience. Do you fix it depending on the political power of the user? Is that right? Does it matter if the user is a physician or a nurse or clerical? If you are in the business of developing software, the answer is easy. Fix it. If you are in the business of supporting an organizations business processes, the answer is not clear at all. Normally the first step is analysis and diagnosis. If it is hours or a couple of days, then I think fixing it is definitely the way to go. If it is weeks, then the answer is never clear at all. I'd like to resolve every issue the user has, doesn't matter the user, doesn't matter the time.

But sometimes work is getting paid to do the things, we don't like to do.