Open-source tool: where your rating actually leaks, from your last 500 games

A local Python tool that pulls your games from the public API and reports where you're losing rating: score by opening and colour, clock left move by move, errors by phase, score after a win versus after a loss, and by rank within a session. Every finding links the games behind it. It also replays each game to find the positions where it's your move and what you actually scored from each — mine surfaced a Caro-Kann position at move 6: twelve games, 12.5%. It then pulls matching puzzles from the public puzzle database, playable in the page via chessground, and ranks studies from the topic pages by likes. The analysis is threshold rules over your own numbers — same games in, same findings out, nothing generated. On top of that there's an optional MCP server, so you can ask an assistant the questions no tab covers ("in my losses as Black in the Caro-Kann under 30 moves, do I lose on time or tactically?" — 26 games, 19 resignations, 2 flags). It only reads the report; it can't invent a number. No token, no account access, MIT. https://github.com/AlexTok09/lichess-analyser Not affiliated with Lichess. Tell me where the thresholds are wrong.