I'm liking the Robert Norton chart about 2/3 of the way down, showing how safety factors need to be adjusted quite radically once we think about how reliable or rickety our estimates might actually be.

Particular kudos to thinking harder about whether we've truly tested the actual environment where our product might be used.

I wish social scientists would do the same in controlled studies of human behavior -- which are then extrapolated to the ways that people make real-world decisions. A particularly vexing examples involves the way that psychology students make decisions in short experiments involving small amounts of money or other rewards. (Endless variations on the "marshmallow test," etc.)

Knowing what a college student will/won't do for a whimsical $5 reward says almost nothing about how an adult on the brink of poverty will balance bigger, more difficult decisions. Yet we apply a 95% confidence level to the college-student experiment and think we've learned something about the power of all financial incentives