I haven't been able to conceptualize a mechanism that incentivises honesty, so the oracle problem really just boils down to a web-of-trust/reputation scheme.

The critical problem is more so that there needs to be some agent receiving bets (as Bitcoin/Lightning transactions) and that same agent redistributes rewards. If there is no centralized agent, then it could be conceptualized as a decentralized p2p betting market, but then there is no forecasting benefit that a prediction market provides, it's just people placing/taking bets at whatever odds they are comfortable with. 

  @straycat is doing some very intriguing WoT work you might find valuable. 

 Once GrapeRank is sufficiently built out, it will be a good tool for curation of oracles