Years ago, when I was making season-long predictions for entire leagues, I would take care to make sure that the projected wins and losses evened out, that a baseball teams that went 92-70 should be counterpredicted by teams that were going to go 70-92. Not that this had to be 1:1 correspondence, but that the totals should get it right.
It was not until years later that I figured out that for bettors, this is not so, and the people with skin in the game understood a simple idea that the rest of us overlooked. The expected over-under of an NFL team is based on average luck for each. Some team might have fewer injuries, an unexpected development in a skill player, or a few key calls go their way, while another might hit the underside of this. Those should even out over the league.
Yet a very few teams will have a catastrophic occurrence: their starting QB might go down for the season in game 2, or three defensive backs get injured in games 3-6. They are then in all likelihood going deeply under their projection. Those are high-impact situations. Big negatives because of injury are much more likely that big positives from avoiding injury. Therefore, the projections for an NBA season of 82 games should not average out to 41-41. The projections should be more along the lines of 44-38 for everyone, even though that doesn't balance in the end. A few teams are going to get clobbered and win 20 fewer games because Anthony Davis or James Harden gets hurt. But in terms of betting, being 20 games below the projection or 1 game below is immaterial. If you bet six teams overall, the other five are likely to gain a game or two and improve their chances of going over the top.