I extend it further: the moment you need to verify that something happened outside the system, your blockchain is broken.

There are ideas around trusted oracles and consensus recommendation ratings for oracles that aren't terrible and may work (no comment on scalability, of course).

It's worked so far for the Dai, a USD tethered stablecoin based on Ethereum. Smart contracts buy or sell Dai against ETH to keep the price around US$1, the oracles feed in the USD/ETH rate from various exchanges. No actual USD are bought or sold by the Dai contracts.

https://coinmarketcap.com/currencies/dai/

Trusted is the opposite of trustless.

The point is that in a trustless system trust can be earned by reputation, which is distributable and relatively trustless. Sure, you are taking the risk that you'll get screwed, but that risk factor should be calculable.

I cannot trust a system that trusts an oracle.

What if it is a trustless Oracle? Augur kind of tries to be this, and appears to do well within a limited scope.

Very limited, i.e. they're down to a few tens of people bothering to vote daily.

The amount of people “voting” seems to be less important than the amount at stake, as if it was a large stake proved incorrectly somebody may come in and fix that. $1.2m trades in 1 market with $800k open interest isn’t that bad.

This is not necessarily true. You can create a marketplace for facts (eg: oracles or prediction markets) and therefore create incentives for bystanders to bring facts to blockchain. Sure, it doesn't have 100% guarantees, but it might improve on the current scenario in certain cases.

Your claim is remarkably hypothetical and lacking in details.

That said, even then - when would it be cheaper just to bribe the voters?

That's why people stake claims, and the more money they put up, the more they bet on what they claim, the more you can be sure that it's not just a troll, not a hoax, not someone trying to move the market, not someone bought and paid for (so not a Sybil attack).

But staking is naturally limited by the amount of money one has -- so it would be a great advantage for big corps with PR budgets and a disadvantage for individuals. Do we really want that?

Hypothetical example:

- The Citizen (on blockchain): EvilCorp is dumping toxic waste on my property! They are making millions of dollars from my ruin! I stake all of my available money -- $20,000 -- on this statement.

- EvilCorp (on blockchain): No we are not. We stake 1% of our discretionary PR budget -- $200,000 -- on this statement.

- PR firm (on blockchain): We stake $200,000 to support EvilCorp. They are our biggest customer, and we can handle the (unlikely) money loss if this means good relationship.

- Other citizens of the town: yeah, that sludge looks really nasty.. but EvilCorp has a ton of money. That guy has almost nothing, I do not think he is going to win. If we support him, we will likely lose our stake. So it makes financial sense to side with EvilCorp.

- Blockchain decides EvilCorp is correct. The Citizen now has backyard full of toxic sludge and broke.

- EvilCorp and PR firm both get even more money.

Good example, and that's why the last few years in block-graph land comprised of going from "just stake it man" to token-curated-registries-of-oracles and other kinds of governance problems modeled and explored. Game theory, handling adversarial edge cases, the good old sharding and securing problem (irrevocable staking to secure "side chains" and whatever Ethereum/Vitalik is working on, but maybe called Casper: https://threadreaderapp.com/thread/1029900695925706753.html )

I have not read this very carefully, but these conversations seem to be worried about classical cryptocurrency attacks -- we have rational agents who try to maximize their money. A threat of someone loosing all of their money will prevent one from doing something.

However, in my example, this is not the case at all. A PR firm or EvilCorp does not care about success of cryptocurrency, they just want to make sure that "The Citizen" will fail. For example, your link talks about "penalizing both sides hard" like it is unavoidable -- but this is exactly what EvilCorp would want.

I don't think I have seen analysis of the prediction markets which assumes actors who are willing to lose money to get the desired results.

That is assuming everyone is moved by money and everything works in perfectly rational way. The problem is - humans are irrational.

No, that's a very bad model. Humans have a hidden utility function and they maximize that. (See also bounded rationality, weak revealed preference axiom for market modeling.)

if you need accurate verification, or it boils down to a single person, agreed. (though that's not often possible, e.g. bribery)

if you need to know the final consensus (or lack thereof) about that verification, regardless of how consensus is determined, it's not a deal-breaker. multi-party signatures, basically, which is what ethereum is designed to do.

(for context tho: broadly I agree, blockchains make zero sense in very nearly all potential scenarios)

That is great.