Component level repair

Of the companies that I have worked for, or been exposed to, all had some sort of feedback loop. Failures were examined and/or logged, the flaws found, and steps taken to correct the flaws. The feedback loops always included the bean counters to some extent, often built in the the budget. The places were I have not seen adequate feedback loops for failure analysis are with inexperienced small startups. When failures are examined, flaws in design or production are often found, but there are plenty of examples of oddball failures that do not give anyone a clue. Often there just is nothing close to a decent sample size to work with, in order to get some sort of conclusion. In these cases - and there are a lot of these cases - blind swapping is the best you can do. For the most part, it works. And as the failures are logged, the sample size grows to something meaningful. -- Will