https://build.moi/@dcm31/when
It's a bit slow because right now it pulls in realtime market data every time.
@AnT Never expected anyone to try to print it! If it were to work, it would here: https://when.val.run/, (but for me the print dialog is just loading for quite some time because it's a really long page.)
@Charlie I also found out that right-clicking inside the frame and printing frame only works too.
Well, I do want an offline copy of the timeline, thus the printing!
This is an amazing project! A minor aspect but I also really like that it tells you what it's actually doing while loading :)
A few ideas on how to expand it even further:
The ability to display just markets from specific topics e.g. just AI stuff, just space stuff, just war stuff etc.
Including markets that predict dates for important developments but were created ad hoc via binary / multiple choice / set markets rather than the dedicated date market type e.g. those listed in the description of this market https://manifold.markets/RemNi/will-we-get-agi-before-2030?r=VGhlQWxsTWVtZWluZ0V5ZQ
@TheAllMemeingEye
Thank you for the feedback!
Yeah it was so slow initially that I needed to see which part was taking so long!
A few people have requested specific topic. Makes a lot of sense! On my next pass, that plus filtering by liquidity/trader count will be my first features.
Regarding non-date markets: It's definitely a bummer that this only uses a somewhat less popular market type. That said, it would honestly be pretty tricky to add non-date markets given that each market has its own internal logic from which to reverse engineer a single date for the timeline. (Actually, one of the hardest parts of this was getting MY calculated dates to match Manifold's dates, even with their source code!).
My takeaway from that would be that date markets need more love, and if there is a non-date market that is actually trying to pin down a date, there should be a date market created for it!
@Charlie Cool, fair enough!
It's also worth noting that for many date markets with options labelled as "[date] to [date]", the final date bucket is used for cases where the event doesn't happen at all, which could distort the output date, since being likely to never happen doesn't mean it's necessarily likely to happen later if it does e.g. something like "when will Donald Trump do a backflip?".
Maybe a way around this could be to exclude markets using said mechanism that have the final bucket at >50% (or other markets with options labelled as "before [date]" where the final bucket is <50%), since arguably in those cases Manifold isn't predicting that it'll happen then, it's predicting that it won't happen.
@Charlie One more thing, it appears the dates output by date markets are the mean rather than median or mode, but depending on context users and laypeople may find the latter 2 more informative since they're not skewed by small probabilities at distant or past time periods
@TheAllMemeingEye Makes sense! To be honest, I would much rather defer to @ian or whoever at Manifold is making the tough decisions about how to distill these markets into a single date. (That's what the existing code does). And if they improve their logic to better approximate things like you pointed out, I would also pray that they would add that as part of the market data in the API so I don't have to rebuild their logic again ๐ฅฒ.
I think in general this timeline is a bit of a love letter to the endeavor that is date markets lol. Like, there's a reason why Manifold has had to bail on them multiple times, but there's also a reason they keep brining them back. It's an extremely intuitive datapoint, but it requires a lot of thought to do right!