Wow, so I guess I was living with the wrong understanding of positive-sum all this time.
Positive sum doesn't just mean win-win. It also includes win-lose because the winning side actually won.
Got it.
Suppose I each year I produce two eggs and you produce two slices of bacon. We both value these at $1 each. The total value produced this year was $4.
Next year, we produce the same, but we agree to a trade - one egg for one slice of bacon. Egg with bacon is clearly superior to just two eggs or just two slices of bacon for breakfast; so we value the combo of egg+bacon at $3, giving total value of $6.
Next year, we do the same. Total value was just $6, for 0% growth. However, that doesn't mean there was a 'winning side' to the trade. It just means the previous estimate for produced value already included the benefits of the trade, so there's been no change.
When your example is included in Norvig's simulation, it will result in two parties coming together, and instead of an interaction of (-1, +1), the interaction (+1, +1) will happen.
However, when such interactions are randomly mixed with zero-sum trades, and the simulaion is run over N (population) and t (time), it will result in net wealth W_after > W_before.
In order to mimic real-world scenario, the prob distribution has to be chosen such that the W_after is between 0.98 to 1.06 of W_before (in other words, between -2% to 6% growth, more or less). Anything far outside this bound would be highly unrealistic for a functioning economy.
However, when you impose this condition, you would realize that the (+1, +1) events have negligible impact, and that (-1, +1) type of events dominate. In other words, Norvig's conclusion pretty much does not change.
And that is my point. Mixing in a fraction of individual positive-sum interactions doesn't change the results, if the total wealth growth is capped based on real-world macroeconomic data.
No, you're missing the point.
0.98 to 1.06, to borrow your numbers, is the ratio of GDP one year to GDP the previous year. It is not the ratio of GDP accounting for positive-sum trades to GDP under only zero-some trades, and falsely equating those is disingenuous at best. The latter ratio might well be an order of magnitude larger.