Usually, a player's top scores are all full combos with good accuracy and similar pp. This is a premise that the generation of recommendations relies on. If the bot says that it can't generate any recommendations for you, it is very likely that your scores do not satisfy this premise. If you have too many top scores which are not full combos + good accuracy or there are significant gaps in the pp distribution, the algorithm won't work. However, if this applies to you, you are in luck. You're still improving rapidly and you will get plenty of pp just by focusing on getting full combos with decent accuracy. You don't need recommendations.
Oh the other case where this happens is of course if you are a top player. In that case you'll have deal with the fact that Tillerinobot is for the people.
That might happen sometime, go to How to fix "confusing-name" error for a solution!
Try !recommend relax. This will use an older version of the system, which has a broader difficulty spectrum.
I got a recommendation with 300 bpm / AR9 DT / CS 5 / some other meta data that I categorically can not play
The recommendations are based on statistics of who-played-what and disregard almost all meta data. This makes most recommendations pretty good right off the bat, but shitty recommendations, no matter how unlikely they are, will always turn up. I invite you to complain about the recommendation by using !complain [some reason]. This is also the basis for a personal complaint, because it logs additional data, which I need to understand how the recommendation came about.
First of all, know that your pp-per-map are weighted to form your overall pp. This means that your overall gain in pp will usually be significantly smaller than the raw pp of the play.
pp per play are determined from your accuracy (acc), longest combo and number of missed notes. The individual contributions differ from beatmap to beatmap. There is a large penalty for misses and combo breaks, so in general players try to get a full combo (FC), i.e. no misses. This is why all displayed pp values assume a FC and focus on acc.
I wrote something about acc vs. combo a while back.
Usually, any inaccuracy should not be visible. There are a couple of things that go wrong from time to time and mess with the calculations. For example, sometimes impossible scores show up (not impossible as in superhuman, but impossible as in more 300s than objects in the map). If you something that is confirmed inaccurate (there is a show which has different pp than what the bot claims), please contact me.
That is a byproduct of the statistical algorithm which produces the recommendations. The value is displayed to give you an idea of the accuracy you should be able to reach.
The bot will probably be more specific in the future. For now, I trust that you can make the map hard enough through your own choice of mods. If you don't feel like it, skip it! You can use !recommend [relax] nomod to exclude mods.
Although I'd be happy to help make your friend list pretty and pink, I do need to keep my friend list usable, so I'll have to decline. Sorry.
Irresponsible Patrons do get a mutual :)
In general, no. If you are looking for maps whose meta data satisfy specific bounds, ppaddict should be all you need. As a thank you for patrons, there are filter perks, see Recommendations.
The recommendations are based on statistics, so top players and players who play with "exotic" mods should not expect anything good. It is hard to say exactly when the recommendations will become meaningless, but there has been good feedback from players as high as top 1k. I'm working on improving those types of recommendations.
No. Although the method should translate just fine, there would be a substantial amount of data to be gathered and processed. Since there are comparatively few players for the other modes and I don't play any of them, I won't be supporting modes other than standard in the near future.
Robot Tillerino and human Tillerino share the account.
No dying has occurred so far. During "rush hour", there can be a lot of messages, but there is a setting in the osu! client, which prevents notifications from popping up while I'm playing, so it doesn't really bother me.
Check @Tillerinobot to see if there is an explanation for this. If there isn't, and the bot remains offline, please contact human Tillerino through every channel available to you!
As a protection against flooding the osu! server with messages after congestion caused by a bad connection (which can trigger the spam filter), the bot may at times ignore requests for a short while and then reconnect. In the following picture you can see the osu servers having some issues, which Tillerinobot detects very quickly and disconnects for a short while to protect itself.
The cause of these issues can be brief disruptions of a variety of services and should not be cause for concern, although lately they've been caused mostly by DDoS attacks. If it persists, check @Tillerinobot to see if there is an explanation for this. If there isn't, please contact human Tillerino.
Also you might wanna check this option 
Not at this time. Please note that while the recommendations are computed in parallel, the responses are not sent in parallel. Responses are sent on a FIFO basis. As a measure against congestion, all input from a user is disregarded while a previous request from the same user is being answered.
The recommendations are not made from scratch when a user queries the bot. They require significant preparation ahead of time and are part of a large number of smaller experiments that I have done over the past months. At this point there is no clean way to publish something that would cover the entirety of the recommendations system.
If you want to support the project, consider becoming a patron:
For more info check out the Donate page!