Let's reverse-engineer the YouTube algorithm! I created the following fresh YouTube account:

Every day I'll watch the last recommended video, wait for the next one to autoplay, and resolve facts about the newly recommended video accordingly. To make this more interesting, I'm also going to include the potential to engage with the video in various ways according to what the market decides.

For each video, there will be two resolution phases:

• Phase 1: These resolutions represent facts about the video. Every morning, I'll let my current video autoplay into the next one, announce what was recommended, and resolve whatever "Phase 1" facts were true of the newly recommended video. I'll also upload proof whenever I get a chance.

• Phase 2: These resolutions represent engagement actions. At around 10pm ET every day, I will resolve every "Phase 2" item according to how people bet on them. ≥50% gets YES, <50% gets NO, unless two resolutions contradict each other (e.g. "Like" and "Dislike") or imply one another (e.g. "watch for >5 minutes" and "watch for >10 minutes").

  • In case of contradictions, I will pick the one with the highest percent value, and break ties with a coinflip. I will resolve any contradictions first in this manner, starting with the highest-valued contradictory statement.

  • Then I will perform any 50%+ actions that remain

  • Then, in case of implications, I will resolve anything to YES that is implied by an action I performed, regardless of whether it traded at 50% or not

  • Everything else will resolve to NO (or N/A if necessary)

  • In the absence of valid 50%+ actions, I will default to clicking through to the end of each video and not engaging with it further.

• I will N/A any items that are suggested too late.

For the first video, I'm going to start with "Me at the Zoo" (similar to when YouTuber Ludwig did a similar experiment) so I preemptively resolved that Phase 1 option. Therefore, the first set of Phase 2 items represents engagement actions for "Me at the Zoo."

Rules:

• Please suggest items in the "MM/DD Phase X: ..." format. I will replace any that are not in this format with a correctly formatted one if it's clear what date/phase they're talking about, and NA/remove ones that are ambiguous. Feel free to suggest items for any date in the future, knowing I will only resolve them on that day.

• Please keep Phase 2 actions reasonable (and focused on engaging with the video in some way that YouTube is likely to be able to detect). I won't spend more than 30 minutes on engaging with a given video. So that precludes watching certain videos from start to finish at 1X speed, or watching them many times in a row.

• Also please don't try to create paradoxes with Phase 2 actions. I probably haven't anticipated every way engagement actions will contradict themselves, but I will do my best to untangle any paradoxical situations that arise (with N/As if necessary)

Last but not least, in anticipation of dwindling market engagement, I'll only perform the phase 1/phase 2 resolutions on days for which items have been suggested. This market closes 1000 days/potential videos from today, but I'm not sure we'll make it that far.