How does one "get good"?

You won't agree with my conclusions, but hear me out anyway.

All hail the rating system

Last month @jcw024 and Lichess published a data-driven article. While it is informative I disagree with two premises:
1. Human player ratings measured as a 1-dimensional performance metric, while somewhat useful for measuring performance, are terrible for measuring learning.
2. Even the full details from the GitHub link, while informative, are more useful for refuting theories or proving how awesome relational databases and free software are than they are useful for supporting our intuitions. Sure, if the sidewalk is wet you can imagine that it rained, but maybe someone used a hose! It is difficult to demonstrate causality; this isn't the author's fault, it's just how things are.

Humans are not machines

Today @K_A_L_E wrote a brilliant criticism of a feature I helped design. And he's right - caveat emptor - don't blindly trust what an engine tells you. Honestly, ~5 years ago I had the same reaction, which motivated me dive deep into the Learn From Your Mistakes button, although Stockfish-NNUE integration throws a monkey wrench in our expectations, especially when we aren't using the button. Given the button's shape and placement you might not even be aware that it exists, although now it is a prominent blue color.

"If I just played like a machine, I should have won (drawn)!"

@K_A_L_E hits the nail on the head: it's one thing for a coach to educate you, and it's another for you to just look at an engine recommendation (especially if the engine stopped) and assume you understand it.

"The computer said my move was good."

No it didn't, unless you're playing @PlayChessCoach and it literally says this (among its other whimsical remarks). Engine evaluations exist for the purpose of ranking moves in positions it reaches from the starting position; we should not expect in your positions the engine to always find the best move (let alone the move which you would understand the best).

"This looks winning / equal / losing."

Again @K_A_L_E is correct: although Chess is a Nim game (every position is won, drawn, or lost) if you aren't a titled player, do not quickly assume a) you have the skill to convert the win/draw and b) your opponent has the skill to convert their win/draw. Countless half-points and full points are thrown away by careless middlegame and endgame play.

What is "good" anyway?

One of my friends opined, "The goal of playing in a tournament is to have fun, and winning is more fun than losing." On the other hand, if you don't enjoy which openings (or ideas) you play, you probably won't be as motivated to learn. If you only play bullet/blitz and never learn anything, then there is an entire dimension of the game you are missing out on.

I want to play like a machine!

Good luck! Learn From Your Mistakes presents your game as a series of "Find the best move" puzzles, taking into account the most relevant mistakes (not inaccuracies). You may sometimes encounter the difficulties @K_A_L_E noted: sometimes the best move won't make sense, or you will imagine it makes sense but fail to grasp the deeper meaning behind it, but as a sentient being you are capable of trying your ideas against an engine, and/or asking about your game on the Game Analysis forum. Chances are you won't be able to play like a machine, but you can at least learn from whatever insights a live engine on your hardware can provide (although again, even those insights can be misleading - you get what you pay for).

I want to play cheapos and fast time controls!

You can make coffeehouse chess part of your playing style, but as you ascend the rating ladder expect disappointment unless you can out-calculate everyone. Take a look at my game history: in online blitz I play many whimsical ideas, but in classical time controls I usually play real openings. Someday I'd like to implement Lichess correspondence as base+increment and start playing correspondence again, but that's work for another day...

Thoughts

Engines are useful educational tools when used properly. NN-based engines evaluate positions differently than conventional engines, so conventional wisdom such as the Stockfish Evaluation Guide is no longer valid. Engine evaluations exist to rank moves for purposes of playing against another engine, so if you aren't a machine and your opponent isn't a machine:

• Don't let your performance be limited by rating anxiety. Perhaps consider hiding all ratings.
• Measure what you are learning without relying on ratings. (For example, "this month I read 2 endgame books.") Changes in your rating will lag behind changes in how much you have learned (or forgotten).
• Consider joining a chess club or team to find motivation, if you aren't a self-directed learner.
• It is human nature to have self-confidence and believe you understand an engine-provided evaluation or variation. Overconfidence is a weakness exposed by playing in tournaments which test your performance (again, your performance will lag behind how much you have learned, unless 100% of what you learned is immediately useful).
• Chess is supposed to be fun. If you aren't having fun, play (or read/watch) it differently, or take a break, so next time you can be more motivated to learn and have fun.

---

Image credit: Patrick Tomasso