How to make a whole game in one day

There are game jams and Flash game competitions seemingly every weekend these days. While I don't really like the idea of pinning my eyes open with matchsticks and coding all night, I do like the idea of getting a complete game finished and released in a single day. I sometimes get a day or two of downtime between client projects, and rather than using this time to experiment on bunnies, I've been trying to work out how feasible it is to release 1 or even 2 games in these gaps.

My first couple of attempts have overrun by about double, and still aren't released (I'll keep you posted), but I have learned a few things along the way that I thought were worth sharing. I've also picked the brains of some other friendly developers for some suggestions (you know who you are, so thanks!)

Here's what I've worked out so far:

• The game can't have more than 1 level that you need to design. So a single maze like Pacman would be ok, but you can't have every level be different.
• You can't require any complex artwork. You need to use either free art, cheap stock art, art you already have kicking around, or art you can put together in 1 or 2 hours.
• You need a game engine to start off with. Flixel and Flashpunk might work for you, but I prefer to use my own engine, which I have been revising and improving for well over a year. It's still nowhere near "finished", but I'm not writing much low-level code for each new game.
• You need a template .fla and/or FlashDevelop template already set up so you don't have to write any boilerplate for Mochi-ads, screen management, preloading etc
• Your game must have a way of automatically progressing - with either randomly generated levels, or a simple way of getting progressively harder automatically. Players expect to go on a journey of increasing peril, but that's hard to "author" in just a few hours, so automating it is the only way to go.
• Your game can't introduce you to any programming techniques that you've never done before - e.g. new type of collision detection. Learning a new technique is a fine way to spend an afternoon, just don't expect a game at the end of it.
• The game must have reasonably similar gameplay to an existing game, so you're not working every aspect of game design out from scratch. If you add even one or two elements to a game that are completely novel, you've done more than most. Combining game mechanics from different games that you enjoy seems like a logical starting point.
• Don't try to make all the code you write reusable - just get the game done, you can always come and copy bits out of it later.
• Don't throw your idea out half-way. Once you come up with your idea and started coding, you might as stick with it and do the best job you can to make it fun. Even if it doesn't turn out to be the new Tetris, at least you will have another completed game under your belt.

If you've got any more suggestions and lessons-learned from your own game-jam experience, please post them in the comments.