> I think the scrubbing/exoneration scenario is straightforward, and the physical anomalies you mention (for the incriminate overlay attack) would be handled by

> I think the scrubbing/exoneration scenario is straightforward, and the physical anomalies you mention (for the incriminate overlay attack) would be handled by CG or 1:1 offline model recording. You could but it would take a lot of effort even if you just wanted to scrub the data of one. You'd also have to know what you're about to scrub before you even do it. You'd also have to make sure there weren't other nodes around too. Keep in mind, every few months the power requirements of the models increase. The next model will have semantic segmentation (https://medium.com/nanonets/how-to-do-image-segmentation-using-deep-learning-c673cc5862ef https://medium.com/nanonets/how-to-do-image-segmentation-usi...) so by Eon 1 things are going to be much, much harder to fake than Eon 0, which we're at now. I'll admit, it's never going to be perfect. But as you mentioned, we can find ways to make it better. But one thing I believe we can guarantee mathematically is that if someone's going to make an attempt to submit fake data into the Merkle Tree, we're going to make them have to pay out the nose just to attempt it. So they have to have a lot of resources at their disposal. But in the end, they're also involved in raising the value of Grassland's currency for everyone else. Because they're funnelling more compute costs into the system, trading their money for Grassland's coin. So everyone else in the network is, in a certain sense, rewarded from the penalty extracted from the attacker's dishonesty. > Maybe the network needs to report on cases where sensors disagree and visualize/surface these n different versions of history. Yes, I'll think about this... Do you have further thoughts on this? > Any thoughts on leveraging something like Golem for the compute and Streamr or Erasure for the payment/sponsorship? I don't really know much about them. Most of the 'blockchain' stuff out there seems quite overly complicated to me and I'm not sure what they even do to be honest (Not to disparage these systems in particular. I think I've only ever heard of Golem before. And the other two I've never heard of). But I'd rather work with systems that I know and aren't going to be changed from under me because of the whims of some other dev team. And Grassland has quite a unique problem space so it was necessary to build the system from the ground up so I know exactly what each piece is doing and what it could potentially do because I don't put in any more code than I need to.