I've worked in the space, and I don't know of any robust method in use that is using Schelling points to solve the Oracle Problem (without the help of any "trusted dispute resolution").

It's done to some degree with prediction markets, but the main problems those generally have are that they predict binary events, which just like binary options are prone to fraud. Even when used for continous ranges (like your temperature example), they generally suffer from semantic "fuzziness" if they are expressed in natural language.