Tactics Training - pairing selection algorithm

Hi all,

I'm interested in how Lichess determines which Tactics Training problem a user will be given. In particular, the rating of the tactic, as it relates to the user's tactics rating.

I have often considered this to be identical to the game pairings, where you're paired with players of a similar rating (plus or minus 100 points). However, I have noticed that the tactics training must have a different pairing algorithm, as problems are chosen wildly inconsistently from the user's rating.

Here are two historical threads about this:

https://socket.lichess.org/forum/lichess-feedback/why-am-i-given-chess-puzzles-500-points-below-my-puzzle-rating

https://socket.lichess.org/forum/lichess-feedback/tactics-trainer-on-lichess-app

I repost these links since there was never resolution or clarity provided. To briefly synopsize, the tactics experience is as though you were a 1900 blitz player and only allowed to play 1200 rated players, which would obviously be unfair and objectively a bug.

In casual hypothetical terms, it would logically seem reasonable to have maybe 50% of the problems be at about your rating, 25% be above, and 25% below. Being generally paired with problems at your level, while sometimes getting problems outside that range to allow for evolution, just like in a normal game. Instead, as you can see from my earlier impassioned posts, there has not been a sensible harmony and the system has felt either rigged or broken by comparison.

I would just like to know what the intended behavior is. Thanks!

2nded, it's annoying to constantly get puzzes 400 points below my rating

But in real games 1600 tactics are important to spot; no use seeing 2300 tactics if you miss the 1600 ones or 1600s will beat you.

On chesstempo.com you can set puzzles to either easy, medium or hard. A similar feature here would be nice

I'm sorry you're suffering the same highly annoying dilemma, but am very appreciative to have this acknowledged and validated by such a higher rated player.

I strongly agree that an Easy/Medium/Hard setting would greatly improve the situation. Based on other threads I read while exploring this issue, it appears this functionality may have existed historically (perhaps only via the website, rather than the app).

I understand the premise of occasionally sprinkling in lower rated problems, however this is definitely not the current behavior: there is a flood of lower rated problems as the default. Again, I'm glad that you agree that the system is broken and incorrect.

I am personally attempting to evolve from 1800 to 1900, and the majority of problems seem to be 1500 or 1600 rated... so I must chip away +4 at a time, hoping to not get one wrong with a, -17, all the while waiting for an opportunity to be offered to solve a problem at the level I've supposedly earned, or higher. I've been surprised how many 2000 level problems I've gotten correct, and wish I could be offered them more frequently than the current maddening ratio.

To be clear, this issue is in its most aggravating and bewildering form only when the user is signed in! If you sign out of your account and do tactics, you will get a huge range made available to you (800 to 2400). Due to the above BUG (or unclarified concept), signing out is virtually the only way I've had access to problems above 2000. So in my effort to learn and evolve I've been using this tactics trainer mostly signed out, which I can live with, but it'd obviously be nice to be signed in to have some idea of my progress.

I very strongly feel this is a bug, rather than a correct design, and am happy to submit this issue elsewhere, if there is a better place for it than this forum.

Is there a better place to vocalize this bug, such as Github?

I don't want to waste more time trying through this forum if the discussion doesn't turn into an actionable, fixable bug with the solution it deserves. This issue has crippled the enjoyment of the tactics and almost gives me no choice but to move to a site that has this reasonably implemented.

I'm sure someone who works for Li is reading this. I try not to get too annoyed with them, after all this whole site is 100% free and I still like the tactics trainer here better than chess.com : Unlimited tactics, all from actual games, easy to access engine to look at said tactics afterwards (not to mention ability look at engine for your games, add variations & annotations & make an accessable study, this site replaced ChessBase for me when my old laptop crashed).

For tactics chesstempo.com is pretty much undisputed #1 (except they don't have easy-access engine like Li). They have premium features but even for free it's pretty awesome.

I'll continue to bump this thread as I agree this is an annoying feature on an otherwise excellent site. A simple EASY/MEDIUM/HARD option would be amazing (EASY = average rating of tactics fed is RATING - 200, MEDIUM = same as rating HARD +200

Do this LiChess & I love you long time & buy another LiChess t-shirt or two. :)

#1 I've avoided commenting on this because I'm curious what other Lichess staff have to say about it.

I did submit a PR which I thought would fix a bug in the puzzle rating system, but there was no easy way to test my patch.

While here I do have one other qualm with LiChess tactics (again stated with much respect for this site!).

I don't like seeing how many times a puzzle has been completed because it pretty much gives away how tough I can expect the puzzle to be.

For instance most puzzles rating 2000-2100 have been (currently) done around 3,000 times. If I see a puzzle has only been played by <1000 members I know it's very difficult.

"I don't want to waste more time trying through this forum if the discussion doesn't turn into an actionable, fixable bug"

I comment to help people with bugs on tactics here but I don't get the feeling there's any attempt to fix them. I'm talking about the fact that different devices/the desktop don't sync up so the scores are different everywhere; little glitches like the last entry in the graph being wrong until you look at it a second time etc. Like someone else said, it's hard to criticize such a great site, but it does make it seem like this forum is a dumping ground the people moan in but nothing gets done. I've seen sites where users post suggestions/bug reports, and other users vote on those ideas so the site owner can see which problems/suggestions are of most interest to people.