People volunteer for 50 imaginary bucks to at some date and time specified and agreed upon by all players, to livestream or not their attempt at making a TAS within one hour, of my specified arbitrary ruleset.
So, for example, within 1 hour, someone should be able to TAS beat Zelda 1, while also burning 13 bushes. That latter condition needs to somehow be programmatically verified.
Another programmatic verification I thought of, was to make sure that any axiomatic rules, (like in Lemmings a player can't move his cursor while it's the other players "turn"), aren't violated. Then, something like Lemmings can be discretized or made turn-based.
I'm thinking, there could be:
a) You submit AI which runs against another AI to see who wins in a multiplayer game.
b) You submit a keypress file in advance, and so does your opponent, and you see who won based on the "blind"/anticipatory keypress file. For Lemmings, or Mortal Kombat. Think about what such a keypress file would look like. Lots of redundant hitting and blocking and building and rebuilding and digging and redigging. I imagine. Maybe I imagine it wrong, a priori.
c) A turn-based competition, where through E-Mail, like chess, each person submits their next Lemmings move. Further rules can be invented, like, a player only has so much time to move, while the other person waits. Or, that a player has to announce their last move, so the other player can continue to play without their intervention. Which is essentially yielding their turn continuously... Until they renege on their "last term" claim?
How can fighting games like MK or SSB be intelligently created in TAS form, to be human in this turn-based respect without past-revision, but to be free of human error like pressing the wrong button or pressing a button at the wrong time. We want to producing Crouching Tiger Hidden Dragon beautiful shit. Not limited by the idea of infinite-undo past-revision which stifles creation of beautiful multiplayer competitive TASes.