Re: Statistics and --rounds

[Gábor Boskovits]

2018-03-03 14:17 GMT+01:00 Gábor Boskovits <address@hidden>:

I will write here a short notice of the statistical implications of building software multiple times in general. I hope you will find this information useful.

Hypothesis: the probability of successfully building the software is greater than or equal to p.

This hypothesis can be accepted with a confidence level c=1-p^(k+1), where k is the number of rounds the software was built, and all the builds were succcesful.

Here is a spreadsheet that shows this in a more useful form. In the spreadsheet you can see, that what is the probability of a successful build give that a rounds n build succeeds and given the confidence level. I used two of the most commonly use confidence levels, but this can be extended if needed.

Build success probabilities.ods
Description: application/vnd.oasis.opendocument.spreadsheet