It wouldn't be too hard, you'd need a model for baseline improvement which should be obtainable by those who do improve slowly and incrementally. Instead of us

It wouldn't be too hard, you'd need a model for baseline improvement which should be obtainable by those who do improve slowly and incrementally. Instead of using a probability density field on the final submissions I would use one based on the likelihood of a person getting to that point within the time they took ~ skill * time. You could ask on submission if it's a worked solution so then you're only looking for false submissions, but since there is no prize for getting a high score there would be little incentive to lie. Those using worked submissions marked as their own for their own ego would have an unusual submission pattern. There are many layout options for the same algorithms and it's likely that a rough percentage of worked solutions could be obtained where a specific layout appears much more often than it should. Worked submissions are also much more likely to be final submissions. Not sure if I remember correctly, but maybe you need to give a solution before progressing. I would make it possible to progress without giving a solution to take away that incentive.