I've been thinking about this a lot recently, observing some problems that people have with exploratory testing.
There is a group of people who don't know at all what to do with an open book. If not given some constraints, they feel paralyzed. They need a support system, like a recipe to get started. The recipe could be to start with specifications. It could be to start with using the system like an end user. It could be using the system by inserting testing values of presumed impact into all the fields you can see. It could be a feature tour. It could really be anything, but this group of people want a limited set of options.
As we start working towards more full form exploratory testing, I find we are often debriefing to discuss what I could do to start. What are my options? What is possible, necessary, even right? There is no absolute answer for that question, but a seemingly endless list of ways to approach testing, and intertwining them creates a dynamic that is hard if not impossible to describe.
I find myself talking about the concept of options that expire. Like when you're testing, there is only one time when you know nothing about the software - before you started any work on it. The only time to truly test with those eyes is then. That option expires as soon as you work with the application, it is no longer available. What do you then do with that rare moment?
My rule is: try different things different times. Sometimes start with the specification. Sometimes start with talking to the dev. Sometimes start with just using it. Sometimes pay attention to how it could work. Sometimes pay attention on how it could fail. Stop then to think about what made this time around different. If nothing else, the intentional change makes you more alert in your exploration.
I've observed other people's rules to be different. Some people always start with the dev to cut through the unnecessary into the developer intent. That is not a worse way or better way, but a different way. Eventually what matters for better or worse is when you stop. Do the options expiring fool you to believe things are different than they are? Make you blind to *relevant* feedback?