Soft consensus is nothing more than a receipt that you are owed something, it isn't in and of itself possession. The truth is that if you don't possess the asset (due to on-chain execution) you might have to sue to get what is yours, and that doesn't always go well.
This is just wrong. It is perfectly possible to create a soft consensus system that is fully autonomous and as with a fully decentralized blockchain, has no one to sue who could possibly help you in moving on-chain assets contrary to the consensus rules.
I prefer the term embedded consensus as soft consensus implies it is somehow weaker or less secure, which is not the case.
I guess (though I haven't really studied them in detail) we have some half-baked soft consensus-ish systems on Steem that are semi-centralized or even fully centralized and this leaves the impression of soft consensus being far less capable than it actually is.
You can have at least 2 types of 2nd layer platforms built on top of steem:
A subchain (that can be a smart contract platform) that has it's own peer to peer consensus rules that only includes operations that are considered irreversable on steem. Such a chain would be slower by nature but the advantage is that any bugs or forks would not have any negative effects on the the main chain.
A second layer where everyone that uses it agrees that if an operation follows certain rules it is considered valid but...there is no peer to peer consensus mechanism built in. Steemnonsters is an example of this type of solution.
The later is not much different than a private transaction conducted between two parties. It's no different than paying someone for a product or service using steem (or any other crypto) but if they don't deliver I have no way of enforcing it on chain (although with steem you could use the escrow function to delay payment until the product or service is delivered).
You could even build some very fancy apps using a combination of escrow, multisig and the hierarchical nature of the key structure of steem along with the secondary interpretation of data that "soft" consensus brings.
Pretty much agree with everything you said. It is a helpful clarification.
Yes and indeed it may indeed be possible to prove this is entirely as powerful as any native smart contracts (for example, with appropriate bonds and penalties to require that participants take the actions which the 2nd layer has determined should be taken).
However, I still argue that allowing simple smart contracts is preferable because that is what many developers want. Give them alternatives with cost, etc. advantages (carrot), not a lecture from the #80 chain on how their whole approach is wrong (stick) and they should abandon everything they think they know about decentralized applications because we (supposedly) know better.