I’ve run into similar problems thinking about how to run payment channel smart contracts on top of plasma. That said, while designing an efficient exit game that works for all smart contracts seems tricky (and I have nothing to offer in that regard), maybe we should try to do it for specific contracts / classes of contracts, and then try to generalize.
- We know plasma cash / MVP work, hence we know “plasmafied payments” work. Similarly any other stateless smart contract should be pretty straightforwardly portable, eg fully collateralized swap options and stateless multisig wallets.
- If you can split the dapp into a small stateful part and a large stateless-UTXO part, you could run the stateful part on ethereum. An example: if there were a trusted on-chain oracle that gives you the result of some real-world event (implemented for e.g. by SGX, and under the security assumption that you trust Intel), you could modify plasma cash to have the ownership changes of UTXOs dependent on the on-chain oracle, which plasmafies the trading functionality of aprediction market. Another example would be a bounty contract where the bounty can be claimed by producing a pair of sha3 collisions; bounty can be funded on the plasma chain but the resolution (i.e. the verification of the collision) done on ethereum.
- For smart contracts with defined participant sets, the construction in State Channels and Plasma Cash can be used, and you do end up getting most of the benefits of plasma out of this, so I do consider this to be a way to “run the dapp on plasma” in some sense. The big reason why you might not consider it to be so is that the bound on how much someone can grief doesn’t change compared to just running your dapp in a state channel.
But it is still an open question to me how far this can be carried, e.g., if an Augur-like resolution mechanism can also be carried out on the plasma chain, since that resolution mechanism does depend on data availability and censorship resistance (e.g. if I see an incorrect resolution I’m supposed to contest it).
Another mechanism the analysis doesn’t rule out is some way to rely on the nonzero consensus strength of the plasma chain, e.g. if it is run by proof-of-stake, presumably the value of the staking coin represents the expected future revenue of transaction fees, and block producers need to have a collective incentive of that amount (e.g. someone bribes them, or they have a position in the prediction market) to censor or to withhold blocks. I think this is something the original plasma paper discusses in the context of nested plasma chains.