IDENTITY LABS / TOOL 01
Scroll to see more
/ 02
THE PROBLEM
A video, a thread, an article and a plain text post are not scored the same way. Each format needs a different weighting of views, dwell and replies before the algorithm decides to carry it further.
So comparing a thread’s likes to a video’s likes measures nothing. It puts two different currencies side by side and calls the result performance.
/ 03
THE BASELINE
The classifier takes a post’s first fifteen minutes as its own baseline, then follows its view rate against that line for the next twelve hours.
Above the line, a post is pulling views faster than it did when it was new. Below it, the rate is falling away — which is what almost every post does. The question is never how many views a post got. It is whether the rate is still climbing.
FIG. 01
Rate vs baseline. Each dot is one post; the dashed line is that post’s own first fifteen minutes. n = 1,284 · polled every 30s.
/ 04
CLASSIFICATION
Every tracked post sits in exactly one of these at any moment. The classifier re-scores on each poll.
VIRAL
Rate is running multiples above the post’s own baseline and still climbing.
EMERGING
Rate has turned back toward baseline. The earliest point at which a post is worth acting on.
DECAYING
Rate is falling away from baseline. The ordinary path for almost every post.
HIGH REACH, DECAYING
A large view count, but the rate has already gone. Looks like a win; is already over.
FLAT
Rate is holding level at baseline without direction either way.
UNSCORED
The first fifteen minutes are not complete, so no baseline exists yet.
/ 05
ENCODING
POSITION
Left to right is the post’s age, from five minutes to twelve hours. Up and down is its view rate against its own baseline.
COLOUR
Greener dots have a newer audience — more of the people watching right now do not follow the author.
SIZE
Larger dots carry more total views.
