950 is actually really close to the 1000 estimate. That "game rating estimate" statistic is really unreliable. The idea was that chess.com could create a feature to give players (particularly beginners not understanding the rating system) to have an idea at how well they played. It's a cool idea, but in application, it's just too difficult to correctly assess estimated playing strength.
TL;DR That statistic is close to meaningless; it's an estimate at best and not always even a close estimate.
https://www.chess.com/analysis/game/computer/352656818/review
according to game review, the 1000 elo bot played like a 950 elo 😭 💀
what.