Thanks for the update, looking forward to the webinar next week!
I think it's important to note though that not all datasets were held out during the chosen out-of-sample (OOS) period, so it's quite possible to have trained factors on this period. For example, all price/volume data, Morningstar Fundamentals, and the two sentiment datasets, were fully available during this period.
Just my opinion, but with the new strategic pivot, I think it would make a lot more sense to have a platform-wide 2-year hold out period for ALL platform supplied datasets. That way OOS for these Challenges would be much closer to 'true' OOS. Yes, one could still get around it by sourcing datasets elsewhere and upload using self-serve, but it would be a much bigger hurdle, including having to train the factor off-platform (if done on-platform, I'm assuming Quantopian would be able to detect if one is accessing the self-serve data during the last 2 years/OOS period), and data cleaning/wrangling, etc.
Controversial maybe, but that's my opinion.
PS: Best would be to use 'live' data accumulated after the Challenge submission deadline (Kaggle-style), but that would mean having to wait 6-12 months for live data accumulate (1 or 2 months would not be enough, especially during these crazy times). Again, just my opinion.