Assess risks for assessing tests

[ke4roh]

Software Reliability Engineering in the style of John Musa. Identify the natural units (warnings) and make a list of the functions (process warning, alert), then break down the cost of failure for each, and figure how bad a failure is in each circumstance.

1. What if it alerts when the storm isn't here? False sheltering - cost proportional to the number of people alerted * $/hour
2. What if it doesn't alert when the storm is here? 2011 (a recent busy year) had 553 deaths from 1894 tornadoes = .29 deaths/storm. Storm track width is .25 mi, warned area might be 10 miles wide, and not all warnings contain tornadoes. Death cost is ~$2M for comparison to the other figures.
3. What if its location is wrong? Is there a way to check the config? Watch for changes in location?
4. What if it alerts when there is no reason? How did that happen? Hacking?
5. What if the software alerts but some hardware fails?
6. What if the power fails?
7. What if the internet fails?