Thread by @Epifrenetics

#NAFO twitter-algorithm informational 2.0: A deeper, geekier dive. Simplified for brain-damaged cartoon dogs.
Main points are still:
1) Replies do not boost a #Fella's tweet directly.
2) Likes, replies, and retweets are *all* essential for broad reach.
3) #RussiaIsANaziState
⬇️

⬇️ 1st, let's make it clear:
The famous boost values are NOT applied to a tweet and don't stack. "Experts" who claim this are mistaking value boundaries for stacking limits and don't understand the algorithm itself.
The boosts are applied to *predictions* of *your* interaction
⬇️

⬇️️I've made this very simplified overview of the algorithm. let's dive in.

Candidate generation is the 1st, crucial step. Currently it's a 50/50 division of tweets from your in-network (people you follow) and out-network (anything people you follow touches).
⬇️

⬇️In-network candidates are selected mainly on historical interaction between you and the tweet author. Having researched the algorithm, a bit, I can add that **visible** engagement is key.

Already here, ~50% of candidate selection is largely based on *mutual engagement*!!
⬇️

⬇️ Other factors play in here, e.g. *language* (bonus for writing in English!), user activity, community ranking, and more that I won't go into.

At this stage, likes, retweets, and replies all count!!

Liking or retweeting is still a huge factor to tweet-candidate selection.
⬇️

⬇️Out-network selection is based on tweet-activity and *estimated topic relevance* (👉 use relevant & trending hashtags!👈)
It uses graph-analysis to estimate how awesome a tweet is, based on the collective engagement of people you follow.
Retweets/likes still essential here!
⬇️

⬇️Liking, retweeting (low effort interaction) and replying to tweets, connects the tweet with new audiences where it can become a tweet candidate.

We are a huge community, but fortunately with low degrees of separation. Let's take advantage of it :)
en.wikipedia.org/wiki/Six_Degre…
⬇️

⬇️The algorithms here are pretty old, and so are the data they've been trained against, as well as the target objectives.

So, we can expect likes and retweets to matter a lot in candidate selection! And ranking doesn't mean Jack unless your tweets get selected as a candidate!
⬇️

⬇️After selection of candidates, comes *Ranking*.
This is where "boost values" apply!

The ranking algorithm compiles a ton of data to estimate whether it is likely that you will interact with the tweet. A higher ranking score = higher probability you'll be shown the tweet
⬇️

⬇️Both historical data + interaction-data from the tweet itself go into this. As does - again - 👉relevant and trending topics/hashtags👈
Whether the tweet author tends to reply back is a HUGE deal!
1st-lvl replies are threaded in, and reply to reply frequency is analyzed, SO:
⬇️

⬇️Having many replies to the main tweet (and replies to these replies) will tend to increase the probability of a higher score.

If the author is seen to reply back, this will itself increase the chances of the massive 75+ point bonus.

But it doesn't go deeper than lvl 2!!
⬇️

⬇️Basically, whatever the algorithms sees your network doing on a tweet, makes it likely to think that you may do the same. This, in turn will increase chances of visibility.

BUT, at *this* stage, likes and retweets almost don't matter at all. Only the replies.
⬇️

⬇️Essentially, liking and retweeting will help a tweet getting selected as a candidate. They're low-effort interaction that DO help a lot!

BUT, it doesn't matter much without some reply history, both on the tweet and in your interaction history!

This is a community effort!
⬇️

⬇️Given what I've found out so far, we can all rest assured that a simple like or retweet still helps these days.
But, we should all make an effort to REPLY w text to causes and people we want to support, for more visibility.
It helps model your feed AND provides a small boost
⬇️

⬇️I'll add: Tweet age and author-community scores are relevant.

Popular Fellas, help smaller Fellas out! It also boosts your own ranking 🫶

Tweets older than a day or two have large visibility penalties - up to minus 40%. So, tweet authors should make new tweets regularly.
⬇️

⬇️This thread was VERY nerdy, so I might soon add a better infographic below here. But I have precious little time to spare.
If you enjoyed it, though, please visit and boost my pinned fundraiser (at the time of writing it's one by @obopollesa ). Drop a dime if you can💙💛
⬇️

@obopollesa ⬇️ I'll add an interesting detail I noticed in the code:
The ranker is trained *separately* on each prediction outcome (e.g. prediction of reply/like/block/follow).

Basically, the algorithm could predict that you will simultaneously follow AND block a tweet author 😂