[im3w1l]

When looking at numerical experiments like this, it's critically important to carefully think about the exact details of the simulation rather than take the result at face value. It's fairly easy to get any result you want by cherry-picking model characteristics.

Just taking a quick look at it, the utility function strikes me as suspicious. No value of freshness? Of course this will penalize youtube who does value freshness.

Another thing:

> they stop reading or keep going based on a probability that depends on the vote they just gave (0% for upvotes, 50% for downvotes, 15% for non-votes)

how is this justified?