The most advanced projects estimator in the galaxy!
------------------------------------------------
Naive - Mean - Median - 60% - 70% - 80% - 90% -
------------------------------------------------
18 - 24 - 19 - 23 - 28 - 35 - 47 -
------------------------------------------------
2.24 ' ' '|'
5.71 ' ' '||||||||||||||||||||||||'
7.42 ' ' '|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||'
8.92 ' ' '||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||'
10.3 ' ' '||||||||||||||||||||||||||||||||||||||||||||'
11.74 ' ' '||||||||||||||||||||'
13.17 ' ' '|||||||||||'
14.7 ' ' '|||||'
16.24 ' ' '|||'
17.87 ' ' '||'
19.62 ' ' '|'
21.65 ' ' '|'
23.76 ' ' '|'
26.34 ' ' '|'
29.51 ' ' '|'
33.32 ' ' '|'
80% ------------
37.81 ' ' '|'
43.68 ' ' '|'
53.4 ' ' '|'
87.69 ' ' '|'
Based on the concepts presented in these articles:
- Evidence Based Scheduling
- Estimate your work better to make that planning conversation a little easier
- How To Estimate Almost Any Software Deliverable in 90 Seconds
Put these in ./myspecs.js
file. You can for example do some data-cleaning before using importing your vectors for example (only use the last n
vectors for example, and brag about machine learning!).
If you don't yet have those, just use [1]
or any list of numbers that you think represent your previous estimation results. For example, 0.5
means you took 2 times as much time as planned, 2
means you did it in half the time.
This is a list of tasks, each having four numbers:
- the most optimistic scenario
- the number you think is most realistic for the task
- the maximum number if the tasks proves to be more complex
- your confidence in your number (between
0-1
) - optional: I add my label for the task, it will not be used by the app
My historical number is about 25%. This is stuff like emails, environment setup, booting in, etc. (Actually it's more like 33%, if I really want to come out of the closet)
The app will first do 100 rounds, where it "corrects" your specs with your estimation history. It will pick a random element from your estimation vectors list, and update (divide the numbers) your specs. Then it will do another 100 rounds of simulation, where it will generate a random number, based on your specs (skewed gaussian distribution).
It will then draw a distribution histogram and present at which certainty how many hours you have to provision.