Granted, I take this hobby more serious than some, but this is my 2c....I approach recipes differently depending on what the expected output is. With specialty beers and specialty meads for example, I often take a proven base recipe and essentially produce the specialty product from it in ~1 gallon increments. On a standard batch this gives me a lot of opportunities to get feedback from friends and clubmembers, competition, etc. before locking down to a certain ratio of ingredients. I do this for coffee beers, vanilla beers, some spice beers, fruit beers, etc. If a beer/mead/cider is going to have some new idea or unproven ingredient or technique introduced to it, in my mind it is even more important that the foundation has been proven and that I am familiar with the expected flavors, etc. Take for example, if you wanted to brew a pecan beer ala some of the pecan beers that are popular in the south, you would want to make sure you were familiar with the base brown beer so that the pecans were the only new variable you were introducing.
To get that initial proven base recipe, I, like most others who have posted, start with an initial research step. I will look in BCS, I will taste classic examples, look at CYBI recipes for classic examples, look up old school recipes that various legends have shared online, etc. Special ingredients or techniques get their own research step as well. Once I come up with a recipe, I will brew it. The very first time I am trying to perfect a recipe I will keep the feedback loop very tight and very aggressively pursue it. When I initially rack that first iteration, I will immediately reuse the yeast to brew the next iteration, and sometimes another varied iteration. I remember spending one weekend brewing 3 batches of English Mild after I had just racked 5 gallons of it into a keg (this probably explains a little bit about my dumping comments in the other thread too). I always only tweak either one thing or two/three very minor things. After I find the iteration or two that I think is the best and most accurately accomplishes what I set out to do with the recipe, I will introduce outside feedback into the process. At that point, it becomes refinement and the pace slows down considerably.
I think the most important parts of recipe development are detailed notes (or very accelerated development schedules like my Mild example above, so that you can simultaneously evaluate different adjustments) and knowing your system well enough that you can not only consistently brew on it but replicate your results. After all, there is no point in finding the perfect recipe if you can never brew it again