Here are rendered MP3s of my practising sessions. Music quality varies...
They're all coded from scratch (synthesis, patterns etc) in ~1hour using plain jitlib. The sourcecode is available here... code
(you might need to copy and paste or select all to see it as I accidentally colorised all the documents)
So rather than listening to the static MP3s above, I'd recommend to download SuperCollider and play [with] the code.
I think I've identified two problems. First the getting-started and then the getting-on-when-you-finally-got-started.
In the beginning there's always the empty document... uhu. So setting up a little task to solve or having an idea of some soundscape I'd like to create seemed to help. It was usually ideas like use fm-synthesis today or make soft drone sounds tonight. Still, many times it started out sounding just awful and I got more and more stressed trying fix it. Any livecoders recognise this feeling? ;-) but pulling some tricks (usually distortion or delay effects :-) and focusing in on a few details - muting other sounds - I think I managed shape all the code to become somewhat okey sounding in the end. I only reset the clock and started all over from scratch twice.
And then when I got nice processes going, it has been hard to let go of that and continue with something different. So the resulting music above is most often just a single 'theme' or process. I feel I'd have to rehears a lot more to be able to do abrupt form changes or to have multiple elements to build up bigger structures over time with. I sort of got stuck in the a of the ABA form. It would've been nice to go ABAC etc. Perhaps that the 1h time limit made you reluctant to start something new 45minutes in as that'd have probably been left hanging.
update 171217: converted all files from .rtf to .scd and attached as a .zip