Inconsistent display of replies

bug
api

#1

Hi,

I’ve spotted a “bug” in the display of replies generated by the API’s POST statuses/update endpoint. For each reply I generate, its display change at random :

  1. the reply is displayed as expected with a notification
  2. the reply is displayed under a “Show More replies” button with a notification
  3. the reply isn’t displayed at all and no notification is sent

In any cases :

  • the actual reply is available in the timeline of the account that created it
  • if logged with the account that created the replies, all the tweets and its replies are displayed as expected
  • the counter of replies is accurate
  • none of the accounts (sender and receiver) have privacy settings issues since sometimes all is displayed as intended
  • the tweet and its reply are public and can be viewed by anyone
  • the replies are generated by the exact same code
  • there is no “sequence” or logic behind the type of display chosen by Twitter

The most annoying part of this bug is that in case 3, the receiver has no way get notified or see his reply, except notice that his tweet has more replies in his replies counter than in the reply list, crawling the “Tweets and replies” section of the sender, spot his own username in one of them and click on it (which one can call “impossible”). So this is a real issue for customer support or games bots for instance.

Here is a sample for each case :

  1. tweet : https://twitter.com/amanodev69/status/1066007507687145472
    reply : https://twitter.com/OuzetSebastien1/status/1066007745843929088

  2. tweet : https://twitter.com/amanodev69/status/1066007484480061440
    reply : https://twitter.com/OuzetSebastien1/status/1066007613484216320

  3. tweet : https://twitter.com/amanodev69/status/1066007527169638400
    reply : https://twitter.com/OuzetSebastien1/status/1066007876907528192

Some topics describe this behaviour, all seems different but I think that the issue is the same, but none of them have a solution :

The most similar topic is the forth (Reply is not linked to the tweet) and I wanted to add some of my informations/investigation/samples, but the topic is closed hence this new one. It talks about a issue in the new interface, but I can confirm that I can see this bug in desktop/web and iOS/Android applications.

Has anyone encountered this issue ?

Thanks a lot.


#4

I have seen this before, and I don’t think is caused by the API (I could be wrong though)

From my observation, there is some algorithm kicking in here when replies appear to be unrelated (or possibly made by an account that behaves like a troll)

This might give you some ideas


#5

Indeed that seems to be related. Thanks for sharing !

It would be nice if a Twitter staff member could confirm this, given my samples (these tweets must be internally flagged “auto-moderated” or whatever) and give clear guidelines to avoid false positive, such as :

  • the account that reply must be mentioned in the original tweet
  • the account that generate the replies must be verified
  • the account that receive the reply must follow the account that reply
  • and so on…

So we bot makers could provide our clients best practices to build their campaigns.


#7

Hi there. Totally understand the angle and questions here - and as a developer myself, I look for predictability in the behaviour of my code, so I can understand your frustration in the behaviours you’re observing. There are a couple of issues with providing 100% clear cut guidelines, specifically that many of our antispam systems that you’re speculating about in this thread are based on adapting to changing behaviours on the platform, and that as the folks who look after the developer platform, we can’t necessarily speak to the ways that Tweets and conversation threads are displayed and presented in the Twitter-owned applications.

The most direct guidelines are provided in the automation rules and policy clarification on automated replies.

There is simply no ironclad way to “assure” or “guarantee” that every single Tweet will always be displayed in a specific context; this may depend on user choices such as hiding Tweets from untrusted profiles, as well as measures to improve the health of the conversation on the Twitter side.

However, if you use the documented APIs in an organic manner and include the relevant parameters (in_reply_to values, with the @handles included or added via metadata) then you should be in the best shape to see predictable behaviour.

We do understand that this may be a disappointing response, but as the platform evolves, sometimes the user interface may move at a different pace to the backend APIs. We thank you for your understanding and patience in building with us.


#8

Hi Andy,

Thanks a lot for your answer ! This will help us to explain to our clients that the display of the automatic replies is not guaranteed.

BR,
Simon


#9

Thank you for your understanding.


#10

At Last an answer! (of sorts :wink:)

It is nearly a whole year since I raised this issue on the forum, and I wish Andy had seen it and replied then, because I’ve been staring at my sh!tty code for a year wondering what I’m doing wrong!

Anyhow, as @amanodev says, it’ll help explain what’s going on to end-users, but it is frustrating to have my otherwise impeccably behaved app (never qts, rts, can only ever mention or reply when an end user asks it a question) suffer at the hands of measures which are supposed to improve the user-experience.

(This moan is not aimed at you Andy, your replies are always helpful and to the point)

best wishes


#11

Again, thank you (and relatedly, sorry for not having responded previously to similar situations); no code is perfect in some situations and weird behaviour is always annoying, I know that both as a consumer of apps, and a writer of code. Our whole team wishes that we had a more clear cut way to answer some of these queries, I promise you! Thank you all as always for working with our API.