Boards›Player Behavior

Fact Checking!

Nightsky Pirate·1/22/2020, 12:32:47 AM·3 votes·7,420 views

*Edited accordingly to responses, thanks for the input guys.

I just wanted to create a list of 'facts' pertaining to in-game behavior and ask that if any are erroneous, please correct them. Examples for clarification would be welcome. The why doesn't matter here, just the facts. The intent is not to open a discussion to opinions on what is good or bad, just the facts. Morality, good strategy and etiquette are irrelevant in this thread. It's not meant as an exhaustive list of all things behavior, but if anyone wants to add to the list, feel free, so long as its with the intent to be sure that your facts are correct, and not to make a player's version of the code of conduct.

  1. Context does not matter at all when it comes to behavior.
  2. Regardless of the degree that a belligerent violates the rules, the player must be reported to be processed by the system.
  3. It subsequently does not matter to what degree a violation is, if enough reports of the violation occurs, the belligerent will be punished. Any degree of "offensive, including language that is unlawful, harmful, threatening, abusive, harassing, defamatory, vulgar, obscene, sexually explicit, or racially, ethically, or otherwise objectionable" (aka negativity) is punishable. *edit: Enough reported games, as multiple reports in one game are equal to one report.
  4. The behavior program is actually two programs; a zero tolerance filter and a dictionary-based machine learning program.
  5. The dictionary-based machine learning program assigns positive and negative values to terms and phrases based on the arbitration of reports and honors. *edit: Not saying honors nullify reports or slow punishments, see responses.
  6. Repeated use of the same phrases and terms in subsequent reported games by an individual receive increasing negative values.
  7. Positive values, compiled in the same way based on honors, can nullify negative values to some extent.
  8. The same program also adds terms and phrases on the same arbitration, both positive and negative.
  9. The same program is incapable of understanding context. *edit: Incorrect- it does identify sarcastic comments to deaths. So far, no other examples of contextual understanding have been shared, so I'd restate it to "The same program does not understand context outside of death times and chat timestamps."
  10. Zero tolerance words are strictly zero tolerance, regardless of language, region or even common use of the term: the program for zero tolerance words being a filter. "MB for not knowing" or "It's just a 320 MB video" is literally as bad as "You n%%%%r c%%t". I remind you that this is only fact seeking. *Edit: Region and language probably matter. "MB" was a red herring, and even the Dutch Kiss ('kys') was overturned from a Zero Tolerance issue.
  11. When it comes to punishments, Riot Support is primarily there to assist players in understanding why they were punished. Riot Support will not help overturn the judgement of the program except in unusual circumstances outside the user's control, such as a hacked account.
  12. A player may select any champion in any role in any game mode.
  13. A player does not have conform to the meta with regard to build, strategy, spells, warding, etc.
  14. A player may not make any build and/or use any strategy- it must bear some semblance to playing to win and not take away from the game from others. Example: six Tear of the Goddess is not approved, nor is Counter-Counter Jungle Support Smite Nunu.
  15. A player does not have to communicate, except when using a strategy that will disrupt the team's strategy. Example: Singed going into enemy base during laning phase to take mid and top minion waves, resulting in many deaths but is ultimately an effective strategy in some situations. *Communicating intent does not override the previous fact.
  16. A player may not deliberately play against their team- intentionally feeding, avoiding team fights, not contributing to objectives, etc. A player technically does not have to cooperate with the team otherwise.
  17. No camps, minions, kills, towers, etc, belong to any role. *Edit: Not true! Leash belongs to jungle, CS belongs to ADC/Mid/Top, generally speaking. Purposefully stealing camps under the JG's nose or CS from laners is a no-no.
  18. Situation and context does not matter. Always play with respect to the rules.

25 Comments

Umbral Regent1/22/2020, 1:27:01 AM8 votes
  1. Context does not matter at all when it comes to behavior.

Correct. There's no situations in which breaking the rules is acceptable, and the context of a situation (whether it be other people misbehaving, a match being a downhill one, etc.) does not matter.

  1. Regardless of the degree that a belligerent violates the rules, the player must be reported to be processed by the system.

Also correct. With the exception of disconnects/AFK's, a report has to be filed in order for the IFS to review and punish their behavior. No report = no review = no punishment.

  1. It subsequently does not matter to what degree a violation is, if enough reports of the violation occurs, the belligerent will be punished. Any degree of "offensive, including language that is unlawful, harmful, threatening, abusive, harassing, defamatory, vulgar, obscene, sexually explicit, or racially, ethically, or otherwise objectionable" (aka negativity) is punishable.

This needs a bit better clarification. Do you mean this being about number of reports in a given match, or that any form of infraction, including minor ones (like report rallying, "gg ez", etc.) can ultimately lead to a punishment when reported?

  1. The behavior program is actually two programs; a zero tolerance filter and a dictionary-based machine learning program.

Correct.

  1. The dictionary-based machine learning program assigns positive and negative values to terms and phrases based on the arbitration of reports and honors.

If by this, you mean "does the IFS weigh number of Honors/Reports into its review", then no. Otherwise, this question can't be reasonably answered, as it would require more insight into the system than is available to us.

  1. Repeated use of the same phrases and terms in subsequent reported games by an individual receive increasing negative values.

As with 5, this question can't reasonably answered. If I had to guess, the negative values of a given phrase would remain the same regardless of how many times it's used or reported for.

  1. Positive values, compiled in the same way based on honors, can nullify negative values to some extent.

No. Honors do not influence reviews in any way, shape, or form. A punishment can still be given for a match in which a player was honored.

  1. The same program also adds terms and phrases on the same arbitration, both positive and negative.

Like with 5, I'm not sure exactly what you mean by "arbitration" here.

  1. The same program is incapable of understanding context.

Since context doesn't matter; yes. The IFS reviews a player's behavior in a vacuum, so it neither sees nor needs context.

  1. Zero tolerance words are strictly zero tolerance, regardless of language, region or even common use of the term, due to the program for zero tolerance words being a filter. "MB for not knowing" or "It's just a 320 MB video" is literally as bad as "You n%%%%r %%%%". I remind you that this is only fact seeking.

From what I know, there might be regional differences in what's considered zero-tolerance, but in general, yes, zero tolerance chat is strictly zero tolerance.

If "MB" is considered zero-tolerance, though, I have to ask; what'd I miss?

  1. When it comes to punishments, Riot Support is primarily there to assist players in understanding why they were punished. Riot Support will not help overturn the judgement of the program except in unusual circumstances outside the user's control, such as a hacked account.

Yes. Players can appeal punishments, but protocol is that if a punishment wasn't issued in error, then it will not be overturned.

  1. A player may select any champion in any role in any game mode.

With some exceptions. Unusual (but reasonable) picks (such as Taric or Leona toplane) don't really need to be cleared with teammates, but more wild off-meta stuff (or things that would directly impact another player's gameplay) need to be discussed with the team, and can be punishable if the team disagrees with the strategy.

Additionally, some Champion Select behaviors can be indicative of trolling, including some Champion/role combinations.

Really, this can't directly be answered as a solid "yes/no". Gameplay misbehavior tends to be like that.

  1. A player does not have conform to the meta with regard to build, strategy, spells, warding, etc.

As above, this isn't something that can be directly answered "yes" or "no". In general, yes, players don't have to conform to the meta, but there will always be trolling behaviors and other quirks that tie into that samesuch behavior that it's not going to be a universally clear thing.

  1. A player may not make any build and/or use any strategy- it must bear some semblance to playing to win and not take away from the game from others. Example: six Tear of the Goddess is not approved, nor is Counter-Counter Jungle Support Smite Nunu.

As with the two above, there's no definite answer to this. Counter-Jungle Support Smite Nunu was a case of the player not discussing their strategy with their teammates or allowing their teammates a say in whether or not they play it. They were punished because they forced their teammates to adapt to their strategy with zero consideration or discussion.

And, to even further muddy the water, do remember that Deathsiege Sion is considered valid and acceptable as a strategy (if it's even still applicable in 2020), and it is very arguable that that strategy bears some semblance to playing to win.

Though, six Tears is a clear-cut example of something not approved, and similar-such behaviors that serve to handicap the team are punishable.

  1. A player does not have to communicate, except when using a strategy that will disrupt the team's strategy. Example: Singed going into enemy base during laning phase to take mid and top minion waves, resulting in many deaths but is ultimately an effective strategy in some situations. *Communicating intent does not override the previous fact.

Correct. Unless you're doing something that will impact your other teammates (such as the Singed base-proxy described above), communication is not mandatory.

  1. A player may not deliberately play against their team- intentionally feeding, avoiding team fights, not contributing to objectives, etc. A player technically does not have to cooperate with the team otherwise.

Correct.

  1. No camps, minions, kills, towers, etc, belong to any role. *Unclear: Does this include a laner intentionally and without approval taking the red/blue buff in a leash?

Intentionally stealing a buff during a leash is punishable if it can be proven as intentional.

As for the main point; Yes to all, and simultaneously conditional no to Jungle Camps and Minions.

As with the gameplay related misbehavior mentioned from 12-14, there are cases in which actively stealing lane minions and jungle camps is punishable; I.E, following the Jungler around and contesting their camps, Support actively CS'ing to steal from the bottom laner, or one lane moving into another's (particularly early on) and actively contesting lane minions.

  1. Situation and context does not matter. Always play with respect to the rules.

This is essentially a rehash of 1, but; yes. The situation doesn't matter and the context doesn't matter; if one breaks the rules, they are subject to punishment.

Silent Gravity1/22/2020, 1:44:16 AM2 votes

(3 Chat is given a ML Score based on it's severity. Things that score higher move your account towards it's next punishment more quickly.

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=A1yJmhb1,comment-id=00010001000000010000000000000001,timestamp=2018-05-31T18:28:22.413+0000)

This is not possible. The reports just flag a game for review. There has to be negative or zero tolerance chat present for be penalized for a chat related offense. We use machine learning to score your chat logs. If there is no chat .. guess what that score is? It's 0.

Number of reports has no bearing on the outcome, aside from needing one report to trigger the IFS to review.

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=iE97lQZW,comment-id=0000000000000000000100000000,timestamp=2017-09-06T22:07:24.834+0000)

Ok, 1 report has the same power as 9.

The machine learning system scores chat. It does not score reports. I believe there is still a lot of confusion about the new system vs the old system. The old system used to work off of report numbers.

The censored word list is in no way linked to the chat log evaluation engine. We let people censor out swearing, but we don't penalize for swearing if you aren't attacking other people or using hate speech.

(4 There any many modules that comprise the IFS. New modules can be created and added as situations arise.

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=3zQ7hXi0,comment-id=00040007000000000000000100000000000000020000,timestamp=2017-11-17T07:50:43.899+0000)

You know about all of them.

The IFS (Instant Feedback System) system calls out to multiple services for evaluation. I should probably write a tech blog on this. There are different services that evaluate verbal harassment, zero tolerance, intentional feeding, afk'ing, etc.

Let's say that tomorrow we needed a new service to evaluate ... morse code. We would write a new service module and register it with the IFS. We simplify the discussion and just usually say IFS. But the reality is there is a very complex group of systems that support player behavior.

and

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=ssrwyraZ,comment-id=0003000100000000,timestamp=2017-12-12T23:40:16.406+0000)

There are numerous detection modules in the IFS, one of which is responsible for detecting a small number of zero-tolerance things such as hate speech. This is the most naive of the modules, since it only looks for things which should never have a place in game. Think of the hate speech module as a first pass detection to easily detect obvious things.

(5 It sounds like you're being intentionally misleading. (6 There is no proof of this. (7 There is no proof of this. (8 The system does learn from reports, but it is manually curated.

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=iE97lQZW,comment-id=00000000000000000001000000000000000000030001,timestamp=2017-09-07T02:19:09.735+0000)

Remember that reports are just used to flag a game for review. So the presence of any reports, would then trigger a review of the game.

The game then goes into our language evaluation service that would return a result for that game. So, I think you question is less about reports and more about how language models are trained. Imperial Pandaa was correct in that the Tribunal was used for the initial training sets. But the system is constantly leaning from numerous sources, reports of course being one of them.

We are also constantly hand checking random samples for accuracy as well as very carefully watching for any cases of false positive. Although false positives are very rare, they do occasionally happen and we work to make sure they do not reoccur. I've overturned a handful of false positives on the boards (maybe 2 or 3?).

If it was a brand new trend, the first time the system saw it, it would likely be discarded. But if that's the situation players exposed to it likely would have no idea what the new trend even was. Trust me, our systems have a much better meme game than you do.

(9 Would an example of someone being punished for 'positive chat' being used sarcastically show that it can understand context?

[{quoted}](name=Riot Tantram,realm=NA,application-id=ZGEFLEUQ,discussion-id=j8OzEwzn,comment-id=0006000000010002,timestamp=2018-08-01T17:52:25.991+0000)

So you got banned for harassing your teammates? That's what the system should do.

Complimenting teammates every time they die is is being negative. You knew what you were doing, they did, and so did the system. The system doesn't look for certain words. It's not looking for 'cuss words' in a list. It's a machine learning system.

Ph03n1xb1rd1/22/2020, 12:12:24 PM2 votes

And where is the sauce for all this?

TrulyBland1/22/2020, 1:23:58 PM2 votes

Regarding these two things:

Context does not matter at all when it comes to behaviour. Situation and context does not matter. Always play with respect to the rules.

There is one common misunderstanding that needs to be cleared up, especially since sometimes even blue posts seem at least badly worded in this regard: The system that automatically detects and punishes infractions is not equivalent to the rules. The system is a tool that helps enforce the rules.

Context does indeed not excuse breaking the rules… but context can show that no rule was actually broken in the first place, and that the punishment issued by the system was a false one.

Very recent and terribly obvious example: https://boards.na.leagueoflegends.com/en/c/player-behavior-moderation/H99Enaga-someone-got-banned-for-inting-because-of-the-sett-sylas-bug-support-said-they-reviewed-it-haha

Outside of context, everything about this look like a pretty clear case of intentional feeding. Within context, it's obvious that it was not.

KFCeytron1/23/2020, 12:39:00 AM2 votes
  1. It subsequently does not matter to what degree a violation is, if enough reports of the violation occurs, the belligerent will be punished. Any degree of "offensive, including language that is unlawful, harmful, threatening, abusive, harassing, defamatory, vulgar, obscene, sexually explicit, or racially, ethically, or otherwise objectionable" (aka negativity) is punishable.

The "enough reports" part should be "enough reported games," as it only takes one report for the system to review a match.

  1. The dictionary-based machine learning program assigns positive and negative values to terms and phrases based on the arbitration of reports and honors.

Valid reports can slow down Honor, but getting Honored doesn't impact punishment status.

  1. The same program is incapable of understanding context.

It has a reasonable understanding of context. Not perfect, of course, but it can tell the difference between a sarcastic "good play" when a teammate dies compared to an actual compliment.

  1. Zero tolerance words are strictly zero tolerance, regardless of language, region or even common use of the term: the program for zero tolerance words being a filter. "MB for not knowing" or "It's just a 320 MB video" is literally as bad as "You n%%%%r c%%t". I remind you that this is only fact seeking.

Perhaps "MB" is part of the ZT filter in China, but it isn't in NA. That said, yes, I've never heard anything to suggest that there are various tiers of ZT phrases. They all have the same detection and punishment rules.

  1. When it comes to punishments, Riot Support is primarily there to assist players in understanding why they were punished. Riot Support will not help overturn the judgement of the program except in unusual circumstances outside the user's control, such as a hacked account.

They also overturn the judgement of the program if it was in error. False positives are rare, but they do happen.

Also, as I'm sure you noticed during the process of creating your post, Boards markdown handles numbered lists automatically. You can make each number a 0 if you want and it'll work just the same. And I'd recommend always creating numbered lists in a program that supports such automation.

Frizzê1/22/2020, 2:53:59 AM1 votes

The fact that IFS has gotten people for things that it shouldn't have in the first place is just another reason to not chat at all.

Posui Gart1/22/2020, 1:40:30 PM1 votes

It subsequently does not matter to what degree a violation is, if enough reports of the violation occurs, the belligerent will be punished

1 report is enough

When it comes to punishments, Riot Support is primarily there to assist players in understanding why they were punished.

I believe there were cases of automatic system being wrong, and support helping revoke wrong punishments

A player may not make any build and/or use any strategy- it must bear some semblance to playing to win and not take away from the game from others. Example: six Tear of the Goddess is not approved, nor is Counter-Counter Jungle Support Smite Nunu.

Sometimes its not as obvious. When Int Sion strategy started, people who used this were banned for inting. Some bans were later revoked, because it turned out this strat is actually playing to win