Floatingdoc;
To put your original plan in a broader context, I like to hearken back to my days as an engineer (long, long ago). All engineers know that, in designing any system, you never assume that there will not be a component failure. Every component has a non-zero probability of failure. The key is to understand the failure probability for every component and understand the consequences of any particular component failure. For mission critical components -- i.e. - the system completely fails if the component fails -- you build in redundancy, because each redundant train reduces the probability of failure exponentially. (e.g. - in a power plant, you have multiple feed pumps running off independent power supplies.) You can also over-design, so that a system can tolerate some level of component failure and still meet the output requirements. (e.g. - design an injection system capable of delivering flow at twice the rate you really need).
Translating to the financial world -- any bond has a non-zero probability of default. It is almost a certainty that in any bond portfolio, you will have some bonds that default. I haven't gone to look it up, but I seem to recall that in normal conditions, the default rate on all corporate bonds ranges between 2% and 4%. If you are investing in high yield bonds, that percentage is almost certainly higher -- they are high yield because they are financially weaker companies. So you have part of your "retirement financial system" that has a probability of failure north of 4%. Others have shown the results of this level of failure on the cash flows of HY bond funds.
There are two responses to this problem. First, you add some redundancy. This is where asset class diversification comes in -- you can suffer a default in the bond portion and still have income flowing from your portfolio. The key, of course is to ensure that your systems are truly redundant (e.g. -- two pumps connected to the same power supply are not redundant in the case of a power loss). This is the problem we saw recently with RMBS backed CDO's, where the designers did not understand that the underlying assets were in fact highly correlated and would all fail together. Second, you can build a system that can survive some level of failure. In other words, have substantially more invested than needed to support your desired cash draw.
What you cannot do, however, is design a system that assumes no failures and barely meets the output requirements when operating properly. It doesn't work in engineering. It doesn't work in personal finance.