Jumping on the chess-cheating-scandal bandwagon and analyzing the ACPL (Average CentiPawn Loss) over time for a selected set of 25+ GMs.
- Games were filtered to only include OTB classical games (no online games, no bullet, blitz or rapid, no simuls, no blindfolds)
- Opening moves (first 10) were filtered out
- Short games (less than 10 moves after the opening moves) were filtered out
- Games were analyzed using Stockfish 15 with depth 15
TLDR;
- Nothing suspicious found with regards to Niemann's ELO vs ACPL.
- Statistical analysis cannot be trusted for a player where the number of games within certain ELO intervals is substantially underrepresented.
Here follows numerous plots of the findings...
As a player's rating is not distributed evenly, but rather in a stepwise manner, it makes sense to group them into tiers (buckets) for certain stastistical analysis. Here I'm using a bucket size representing 50 ELO points.
