Hi Haplo,
Thanks for letting us know about your feelings.
The bot ratings have not been manually changed for years. So I am not sure why you feel this is the case.
Bot ratings do evolve like human player ones though. So they will fluctuate. The evolution speed, however, is much much slower as the bots play lots and lots of games and their skill do not vary with time (they don't make progress unlike humans) so we picked a much lower coefficient for them. They will only lose/win 1-2 points max per game.
So I'd say bots have a more accurate rating than humans actually.
When a new game is released we seed the bot ratings according to our series of tests (they play hundreds of games against each other), but this is only a guess as humans will play differently than the bot opponents and also some bot weaknesses might be more or less exploitable by humans (so a bot we think is stronger than another one may actually be weaker). We try to have the 12 bots in order of strength so we may from time to time adjust a few parameters up and down to try and force them back in order of rating. But that's certainly not what you are seeing.
After the initial seeding, the bot ratings will definitely move up or down, depending on how strong they really are against the human players. I think in Hanamikoji, the bots were initially seeded too high. So their ratings have sunk down quite a bit since the release of the game.
I am pretty happy to hear that a top 10% player can still be consistently be beaten by bots. It shows that we have worked well on the AI side. You may be used to it on Happy Meeple but it is far from being a general rule. I have seen countless apps where AIs are pretty bad, even the supposedly strong level.
I hope this answers your questions.
PS: a couple of months ago, we improved the Lost Cities bots a bit, but according to our tests, the improvement was only 30 points (and up not down).