Bad forecasts

[walles]

Looking at this pic it's pretty obvious to me that when each number takes / can take a long time, that increases the uncertainty of the forecast, but we don't take that into account.

I think:

• the projection should have two parallel lines
• the X distance between the lines should be related how long any single number is expected to take
• the slant of the two lines should maybe be calculated using least-squares?
• When we have multiple observations for the same number, maybe keep the first and the last of those and ditch the middle ones so we don't bias ourselves to when lots of observations were added?