This post originated from an RSS feed registered with Java Buzz by Michael Cote. Original Post: Yesterday's Weather Feed Title: Cote's Weblog: Coding, Austin, etc. Feed URL: https://cote.io/feed/ Feed Description: Using Java to get to the ideal state. Latest Java Buzz Posts Latest Java Buzz Posts by Michael Cote Latest Posts From Cote's Weblog: Coding, Austin, etc.

Brandon writes that doing projections/estimates seems silly. They're almost always wrong. Instead:

All that matters is what you have done. The actuals. How much you made last month. How much code is written, how many task are complete. It is the only thing that is accurate. I wish we could take all the work on projections and redirect it into finishing the job at hand. Oh what a world that would be...

To which Lunt replies (in the comments):

If I ask someone how far it is from Austin to Dallas, and they say, "I don't know", I'm pissed. Is it 30 minutes, less than 5 hours, more than 100 hours? I want some kind of projection, even if it's 100% wrong.

Driving projectionss are easy 'cause you're doing the same thing over and over again (I've driven to Dallas countless times, and it's always the same within 1/2 hour...except that one time I got a flat), with almost 100% known tools (car, road, etc.).

But, if you were to project how long it would take you to get through unknown territory (like developing/innovating new software), I imagine you could come up with a rough estimate, but not a very good one.

Along those lines, I tend to favor the "Yesterday's Weather" theory: instead of estimating how long something will take, just figure you'll get done today as much work as you got done yesterday. This entails keeping track of your past performance, of course, which is a lot more than most planning and estimating schemes seem to end up doing.

So, if I'm going through uncharted lands, and I was able to move 10 miles yesterday, I'll say that I'll probably be able to move 10 more miles tomorrow. Then if I actually move 12 miles, the next day, I'll figure I'll move 12 more. And so on. It'll never be 100% accurate, but it'll be closer than making stuff up.

There's a problem of boot-strapping this whole process, and as such, it'll be very wrong in the beginning. But, as more and more history is built up, it'll get more and more accurate. It'll never work out perfectly, but as Fowler says, "The point of course is that while Yesterdays Weather is a crude mechanism, it ends up being not significantly less accurate than more sophisticated (i.e. complicated) ways of doing it."