For the past few hours I've been struggling to find an idea that satisfied me. I've had ideas--I posted some of them in my previous journal, I believe--but I've found problems with all: some didn't seem to use the theme well enough, others seemed to require grinding, a few just seemed insipid, and so on.
However, I think that I now have two concepts that might work, and have only to decide between them before I start developing a prototype:
First is a design mentioned in my previous post: the player controls a set of cultists attempting to perform a ritual sacrifice at several particular locations, while avoiding heroes intent on stopping them. At its heart this would be a stealth game: the player attempts to reach the ritual site without being spotted by either civilians (who might raise a "suspicion" meter) or guards and heroes (who would kill the character, and who are alerted once the suspicion meter becomes full). Death would be useful in two manners: first, performing the sacrifices advances the game, and second, being killed by a hero "outs" that hero, making them visible from the outset of subsequent attempts.
I really like this idea, but I'll confess that I have qualms about making a game in which one plays a set of villain-protagonists of this sort. To some degree this might be ameliorated by making the sacrifices be ritual suicide, removing the element of murdering others, and by the fact that I would intend to portray the outcome of completing the game as something very unpleasant: "winning" would produce a screen that depicts the horrible arrival of the being summoned by the ritual, while losing would announce that the world is safe from the player.
My second idea is something new. The theme is that "death is useful"--but it doesn't say to whom death is useful; in this concept, player-deaths are useful to the enemy. Specifically, this game would be a procedural shooter with perma-death. When the player dies, their most-successful (or most-used, perhaps) weapons are distributed amongst the enemies, and a boss based on the player. (This is likely to be somewhat weighted according to the player's success in the run: a player that did poorly might end up "donating" only one or two weapons, and might not be made a boss.) Furthermore, enemies that performed poorly against the player receive a lower weight during the generation of the next run.
Again, I rather like this idea. On the other hand, my submission last year was a procedurally-generated game, and that took long enough, and was judged poorly enough, that I'm hesitant to follow so similar a route this year. On top of that, there's a danger in using an idea that just recently came to me: I've found in the past (including last year's submission) that it can take time for me to spot the problems with a concept; I recall that this was, after all, part of the reason that I wanted to dedicate a day to concept-work.
I made a game once where enemy deaths were still useful to the enemy, but only in the sense that they fell out the sky but continued to absorb your shots! The idea you talk about here seems more like an evolutionary process, where death is indeed useful, natural selection and all that.