Jon Pretty

I'm glad @fulhack wrote this clear analysis about estimating software projects. It maybe needs another explanation about why the mean and median completion time for software development tasks are so different, so here's my take. (1/4)https://twitter.com/fulhack/status/1117968632536010754 …

Unlike almost any other production task or industry, software development seeks solutions which are, at their most basic, repeatable; at best, highly abstract and general. One solution can be reused thousands or billions of times for almost zero cost. (2/4)

All the tasks which can be completed using known solutions, no matter how large they seem, will be competed quickly. So what's left after that? It might be a small part of the total, but it will need new solutions, working things out for the first time; all the unknowns. (3/4)

So that's where almost all the time is spent: tasks that you've never solved before. And how do you estimate them, when you have no data? It's at best a guess. (4/4)

No different than my experience in (big, many $$ millions) construction projects. I would have to estimate schedules/costs/materials with little data/requirements then do it. There is a rich literature on time/cost overruns in complex projects and software is no snowflake here

But construction projects don't have the feature that the majority of the perceived work is basically free, as it often is in software. Even the very worst overrun construction projects, like Berlin-Brandenburg Airport, aren't more that 5x over schedule...