They are aiming for 6 launches this year. That is a lot of chances to make it work.

Even if there is a catastrophic failure that destroys the launch tower, they are building another and will still likely get 2-3 attempts.

Bullish? Is there someplace I can bet against them? I don't see how Starship isn't getting into orbit on this or the next launch, barring major regulatory sabotage.

How does Metaculus make such predictions? I assume it can't be in the reliability engineering fashion used in industry.

They make it up.

But everyone assumes it means something because it's an average of a bunch of people making it up.

No need to assume, you can check the prediction history and see that they generally do quite well. You also look at the histories of individual predictors' profiles and see some with consistent successes.

Well what is interesting is not how well the aggregate does, but how much better it does than individuals. After all if individuals can predict stuff consistently then what does the prediction market add?

My understanding is that the market, which rewards and punishes your predictions is what makes one a better predictor.

Right, but that doesn't do anything unless there's life-savings-amounts of money on the line such that anyone is incentivized to actually try. And even when people are incentivized, just look at the stock market for clear evidence that it doesn't work: a market doesn't magically make people good at things; people screw up and lose bets and go bankrupt all the time. Sure some of the decently-competent banks and hedge funds don't---but there is no reason to think there are any actors close to that sophisticated on a prediction market.

You're of course that it's not yet as sophisticated as the stock exchange, but I wouldn't be as dismissive of the importance people attach to internet points.

The problem isn't that they don't place importance in them. It's that there is nothing weeding out failure. Financial incentives in principle (although definitely not very well in practice lol) punish failure in a resilient way: you actually can't keep betting if you can't win. Where... I don't think that's true for prediction markets?

Anyway, I dunno. I also don't believe that people predict well even with incentives, and think the whole thing is a farce, so I'm just a pretty dogged skeptic.

Unclear, but the predictions appear to be better on things like US elections than prediction markets or 538: https://firstsigma.substack.com/p/midterm-elections-forecast...

Wisdom of the crowds.

How well does this work when there is fundamentally little information? It works in stock trading because so much information is standardized and regulated. I am skeptical of it being as useful here simply because the crowds don't have much relevant information, as far as I know. For example, do the crowds know about the design redundancy? What about the supplier quality control? Or the reliability data on the check valves used? etc. etc.

Engineering, physical principles, I guess. Those who place bets or making predictions may know a thing or two on mechanics, rocket science and history of building launchers.

To put a finer point on it, those seem like rather blunt instruments for prediction compared to how risk is typically measured by aerospace reliability engineers.

Think of it like top down budget va bottom up budget.

RAMS engineers do a bottom up analysis. However, it is fairly typical for the to be subtlety biased in a way that changes the final answer. (Usually by ‘rounding up’ on individual hazards).

Top down takes an outside view and says ‘this looks like these other past projects. It should land in around the same place.’

I’ve found top down tends to be more accurate project wise. Bottom up more accurate for specific subsystems.

Related to this discussion, the results found by Feynman in the Challenger investigation found that the top-down risk assessment was off by orders of magnitude.

I'm sure some aerospace professionals are wondering how it's possible to predict stock behavior on an exchange, but have no problem with educated predictions of launcher performance.

That right. The point being, they are very different systems with different information inputs. Meaning what works for one may not be a good method the the other.

but also skin in the game and supply/demand

Should be much more accurate than a simple poll, because people with the belief that the odds are undervalued will put money in the game to hopefully profit off their belief

It's more like the Efficient Market Hypothesis

There's no skin in the game with Metaculus, it's a prediction aggregator, not a prediction market. Nothing is at stake.

This is why you see a somewhat absurd skew on AI doom/safety questions, using Metaculus is a hobby of the cult^H^H^H^H movement, and no one's swooping in to steal their money because there isn't any.

It’s somewhat more complex than that because Metaculus weights its aggregates by how accurate predictors have been in the past. Definitely not perfect though, and the efficient market hypothesis isn’t relevant at all.