Turning this around puts the negative outcome, drowning, in the forefront. Including specific reasons might help, but it is easy for the student not to internalize these explanations, especially since they would likely prefer to not think about drowning. Telling someone not to drink on a boat because being drunk will make it easier for them to drown is too vague. Telling someone to not drink on a boat is a prescriptive rule. Most people are more likely to follow rules they understand (so long as they agree with why they exist). I think one of the reasons why this works so well is that it makes the student think about the 'why'. But it's one analysis tool, and it's worth being careful of applying it to the exclusion of other ways of analysis that lead more directly to the synthesis of the 'correct' answer. I don't mean to attack a strawman version of this way of thinking. In other words, sometimes it's easier to enumerate goodness than badness. This is where you need, again, a simplifying insight (like 'look from the outside' or 'use a quorum' or whatever). Compare your answers, and you'll likely find their list of badness is different from yours.

What are all the ways a server can suffer gray failure? Spend a minute writing them down (bad RAM, bad RAM that still passes ECC, CPU corrupting floating point calculations, etc etc). Then there's the issue with enumerating badness. It's a brilliant insight, which doesn't seem to lead directly from the set of problems. In the case of Paxos, that's having acceptors return their previously accepted values and having proposers drive those values instead of their own proposals. So I can churn through loads of broken protocols without identifying the key insight that leads to a working protocol.

That finds a lot of problems, but doesn't necessarily find the solution: the space of problems is dense, but the set of solutions is small and tightly clustered in a few 'islands'. Then find the next problem, and close that, and so on. So I start with things that don't work, like 'fire the packets off at each server and hope they arrive in the order I sent them'. Say I'm designing a distributed replication protocol, basically trying to discover Paxos from first principles. However, it does have two downsides as a way of engineering thinking that is worth being aware of: completely enumerating badness is hard, and the solution space might be much sparser than the problem space.

I think I also learned it from Charlie Mungers book, but I don't recall clearly. This is a very powerful way of thinking, and one I use often when thinking about things like career direction and personal life.

Prison architect can't start programs