I will copy several comments I have made before elsewhere.

Firstly: "Non predictive" is just a name for a category of "markets where profits shouldn't be counted towards some things". The "non-predictive" tag (currently) has two effects: Manifold doesn't pay out author bonuses for them, and Manifold doesn't count them in leagues. It is important to remember that the point is to reward markets that Manifold thinks are worth rewarding, and not reward ones that Manifold doesn't want to reward; and to count predictive score on "real" predictions rather than e.g. winning self-referential games or collecting bounties. The name is just a name and should be changed as needed to suite Manifold's goals. ("The Categories Were Made For Manifold, Not Manifold For The Categories")

Now, on to the topic of what do I think best reflects what markets Manifold wants to reward making and betting on?

I was thinking: Currently, self-referential markets aren't counted in profits, because they don't predict something external to themselves and that's something Manifold has decided should not be rewarded as much. There's also been a lot of discussion about whether markets about "How much profit will Alice make?" or "Will Bob win their league?" and whether their profits should be counted. Then there's questions like "Will Manifold user Charlie tweet today?" where Charlie is actively betting on the market and can decide whether to tweet or not to decide the outcome.

I propose that the criteria we want for whether to count profits is a more generalized form of "Market resolves based on its own activity". I think probably the criteria we're looking for is "Market resolution is substantially influenced by activity on the market itself".

• "Will this market get at least 10 traders?" - obviously counts

• "How much profit will Alice make?" - if Alice is betting on the market a lot then it will influence their profits and therefore the resolution. If Alice isn't betting on the market then it doesn't. (Since there's no way to prevent Alice from trading, perhaps you don't count profits on these markets by default, or perhaps you see whether they did trade or not)

• "Will Bob win their league?" Same as previous example

• "Will Charlie tweet today?" or "Will Charlie do 10 pushups?" Same, if Charlie is betting a lot, the market is quite likely to be a big influence on whether Charlie does the thing and therefore on how the market resolves.

  • Note: of course, these markets are often good! They can be used as motivation or bounty markets to accomplish things. But clearly the profits made on them, at least by Charlie, tend not to reflect predictive score. And sometimes they are good, while sometimes they cause bad incentives and harmful outcomes.

Of course, the phrase "substantially influences" is doing a lot of heavy lifting there, and it's not clear how to define it. But I argue that any criteria you try to set - even the ones we have now - are quite ambiguous and ultimately you have to make judgement calls. What it means for a market to "substantially" influence its own resolution is unclear, but I think the core philosophy of this criteria makes sense.

The basic point is that self-referential loops create incentives for people to manipulate the outcome for profit, and as the stakes get higher the incentives become more likely to be more problematic. There's nothing special here about markets that directly resolve on their own activity - a wide variety of markets can become mostly self-referential when the stakes become high enough, even if they are mostly non-self-referential when the stakes are low.

I should also point out, this criteria may be very hard to enforce, and by taking away rewards when a market appears to be substantially influencing its own resolution, you would also potentially incentivize people to hide that and be more deceptive. These are difficult tradeoffs, and I don't know what's the best way to address them.

Made a market for if they'll change markets about leagues counting for leagues:

I don't know if I fully endorse any of the options listed above so I voted for case-by-case basis. I think one interesting tension in my beliefs is "what I think it is fair to reward people for with league profit/promotion" and "what I think it is fair to punish people for with league losses/demotion."

I wonder if there would be some merit to some markets having their profits/losses count for leagues fractionally, while still accruing trader bonuses. I feel like "silly markets" like the Liz Tweet ones might have been better if all the mana gained/lost counted for leagues, but only at a 1:10 ratio or something. The same might also be an interesting way to deal with markets about leagues itself.

These ideas might be bad, but I know David is going to be thinking over the problem so It's good for us to get all our ideas out there for him to see and hopefully some of them are good.

Also, I think that the rules should ideally be clear enough that any trustworthy-ish user can be given mod privelages to enforce them so it's not just david and the auto-tagger.

@Joshua I probably should have included an "Other" option for this, but I thought it would funnier to have the punch line as the last option.

@Joshua I don't think there's any problem with trader bonuses from Leagues markets counting for Leagues, but maybe profits from those markets shouldn't count, since that can make them self-fulfilling or self-defeating.

@Joshua Yeah, I definitely think the rules should be clear, which is why I don't think that doing it on a case by case is a good idea. I wouldn't be happy if I had a big, popular market that I thought was predictive, but then it suddenly got marked non-predictive, or if I had lost ranks in Leagues after a market I profited on got marked non-predictive.

To me, "non-predictive" means "this market doesn't predict anything about anything other than itself." For example: "will this market get 50 traders?" is non-predictive; "will the LK-99 market get at least 5,000 traders?" is predictive.

EDIT: An example of a market I've made that's very silly/memey but also still predictive: this one about Vivek Ramaswamy meeting the same fate as Vivec City. Silly, yeah - but I'm also actually going to resolve it as YES if it happens, assuming the asteroid/moonlet isn't large enough to end multicellular life on Earth.

@evergreenemily Something I didn't think to put on the poll is whether markets about non-predictive markets should be counted as non-predictive themselves. I don't think they are non-predictive, since they are definitely predicting something that isn't about themselves, and in many cases, they don't even influence the events being predictive, but some that I have made were marked non-predictive, e.g., /JosephNoonan/will-the-burger-blast-market-be-dec

@JosephNoonan I would count those as predictive, personally, but I can see why others wouldn't.

tbh I still think non-predictive shouldn't be a thing. the whole problem is the interaction with leagues, not the gambling

@Stralor I don't have a problem in principle with removing certain markets from Leagues, but I think there should be clear standards for what markets are excluded. Making a category for non-predictive markets and excluding them could have worked, but it seems like the category of "non-predictive" keeps being broadened to include any market that people complain about on Discord.

@JosephNoonan agreed. and the punishments are so steep that it feels like a fine

@Stralor It pains me to see how much I've lost in UTBs every time I look at this market: https://manifold.markets/JosephNoonan/freeresponse-almost-freeforall-whic?r=Sm9zZXBoTm9vbmFu

But at least that's a definitively non-predictive one. Although it wasn't marked as such until I mentioned it to the admins, so I could've gotten away with so much extra mana if I wanted to.

It understandably feels like a punishment for a market to get marked as non-predictive, but I would suggest that a better framing is that Manifold rewards select markets that they deem worth rewarding, and doesn't reward the rest. The selected markets today happen to be most markets, the ones called "predictive".

I think Manifold could make this feel better by flipping the labelling around - instead of having the "non-predictive" label "added", you could "add" the "predictive" label to pay out author rewards for them.