Follow doesn't mean care: the fiction of the following list
Why follower counts and following lists overstate real bonds, how interaction evidence corrects the fiction, and what a directory built from measured exchanges shows instead.
A follow is a one-time, costless, effectively irreversible social gesture -
and your brain reads it as a bond. The archive separates the two things: the
graph (who follows whom, from the connections snapshot) and the evidence
(messages, reactions, searches, story activity over time). Every view that
matters here joins them: you can see who is on the list, and you can see who
has actually done anything with you in the last two years.
"1,847 following" is an inventory of clicks, not of people. Nobody has 1,847
relationships; sociological work has shown for decades that meaningful social
circles are small and layered - Dunbar's work puts active, invested circles
in the low dozens - while the weak-tie layer beyond them is real but thin.1
The following list mixes all of it into one integer: your closest friend, a
restaurant you visited once in 2019, an account you followed ironically, a
person you no longer remember, and several that no longer exist.
The psychological error is treating a binary, zero-effort, one-time act as
if it were an ongoing, effortful, reciprocal state. Your export lets you
see both layers side by side and stop conflating them.
The graph: what a follow actually was
In the connections files, the platform hands you the raw sets: followers,
following, close friends, favorites, blocked, restricted, muted, pending and
received requests, recently unfollowed - each a list of handles.2
The product indexes them into one per-account structure, so any single person
can be checked against every list at once:3
- Are you following them? Are they following you? Both? Neither?
- Are they close friends / favorites - *and what did those lists mean in
usage*? (Close friends gets a story typed for it; a favorite may never have
received a message.) - Are they muted - the graph's own admission that presence outpaced interest?
Each fact is true, and each is weak evidence of a relationship on its own. The
list is the platform's model of connection: cheap to enter, cheap to stay in,
and optimised for distribution (who sees your stories) rather than for
describing how much two people actually talk.
The evidence: what happened between you
Alongside the graph, the export carries behaviour with timestamps:
- Messages - who wrote, how often, how recently, how fast replies came.
- Reactions and comments - the acted-on layer.
- Searches - the attention layer you never performed for anyone.
- Story activity - what you viewed, what you reacted to.
From these, the analytics build what the graph cannot: a per-person strength
tier over measured exchange - inner/active/fading on the strength side, and
core/active/at-risk/casual/dormant on the engagement-intel side - where
at-risk is not a feeling but arithmetic: their recent reply rate dropped below
30% of their prior rate, past 21 days of silence.4 Fading is the
timewise version: a tie that exists on paper, stale beyond the 180-day
cutoff.5
None of this rewrites the graph. A person can sit at the top of your following
list with zero messages and a decade of story views; that is real too. The
point of holding both layers is that the word "friend" stops being load-bearing
in either direction - you get a list, and you get evidence, and you choose.
Why the correction matters psychologically
Three familiar distortions follow from conflating the layers:
- Ambient guilt. People appear in your following list that you have no
active relationship with, and the presence alone - unexplained - reads as
an obligation. Naming the layer ("they are a 2019 restaurant, not a
person I owe") dissolves the obligation without any action. - Miscalibrated trust in size. A big graph feels like a support network.
Support is delivered by the inner layer, which the archive can actually
count - and which is usually an order of magnitude smaller than the
follower number.1 - Misreading silence. Someone on the graph who has gone quiet may be a
fading tie (stale by measurement) rather than an enemy (a story your mind
supplies). The 180-day cutoff exists to label the first reading; the app
never offers the second.
Where to see it in the app
- Accounts directory - filter and sort by relationship state, tier, list
depth, activity presence; the columns show graph state and evidence state
next to each other.7 - Per-account page - every list the person is on, their tier, their last
exchange with you, your searches for them.8
- Relationship views - the strength tiers with their populations, the
stale-tie and re-engagement sets computed from real dates.9
All of it is derived in the browser from your ZIP. The directory's sorting is
a reading order, not a hierarchy of worth - the same refusal that keeps scores
out of the product keeps rankings-of-people out of it too.
Should I unfollow everyone I don't talk to?
The archive does not say, and this post won't either. Weak ties have documented
value, lists are cheap to hold, and pruning is a gesture with its own social
consequences. What the tool offers is accuracy: which layer each person is
actually on, so any decision is made with both the graph and the evidence in
view - not on a hunch about a single integer.
Does a high follower count mean a strong network?
It means many one-time follows pointed at you. Whether any of them involves
reciprocal exchange is a separate question the archive can answer per person -
message exchanges, reply rates and recency - but never by inflating the count
itself into a strength claim.
Someone follows me but we never speak - are they a real connection?
They are a real graph entry. Whether they are a relationship depends on
evidence the file carries on both sides: story views, reactions, searches,
absence. The tier vocabulary here (silent-fan, dormant, fading) exists to
describe exactly that space between "on the list" and "active with you"
without pretending to know what they intend.
Why does 'at-risk' appear in the tier labels - is the app predicting my friendships?
At-risk is arithmetic over past behaviour: their recent reply rate fell below
30% of their prior rate and it has been over three weeks since their last
reply. It describes what already happened, measured from message rows. No
forecast of what happens next is computed anywhere - and the Compare view
states its own copy disclaims prediction.
Related reading
The relationships post for the full inventory of connection files and what each
list means; the finding-people post for how the directory turns those lists
into lookup and filtering; the parasocial post for the direction-of-attention
side of the same question.
1: Dunbar, R. - social circle size research (neocognitive limits on
maintained relationships, layered 5/15/50/150-style structure); cited as
research framing, not as a claim measured by this product.
2: packages/shared/src/analytics/archive-compare.ts:9-32 -ArchiveSnapshot fields: followers, following, closeFriends, favorites,
blocked, restricted, muted, pendingRequests, receivedRequests,
recentlyUnfollowed.
3: packages/shared/src/analytics/user-index.ts:3 - RELATION_KEYS
covering followers, following, unfollowers, fans, mutuals, closeFriends,
favorites, blocked, restricted, muted, pendingRequests, receivedRequests,
recentlyUnfollowed and more, all indexed per normalized handle.
4: packages/shared/src/analytics/engagement-intel.ts:173-176 - theat-risk rule (score >= 35, recent reply rate < 0.3x prior, > 21 days since
last reply) and tier thresholds core >= 60, active >= 35, casual >= 15,
else dormant.
5: packages/shared/src/analytics/relationship-insights.ts:10-12 -STALE_DAYS = 180, VERY_STALE_DAYS = 365; tier assignment atrelationship-insights.ts:45-49 (inner / fading / ghost).
6: Granovetter, M. - "The Strength of Weak Ties" (1973); cited as
research framing on the instrumental value of distant contacts.
7: apps/web-next/src/views/Accounts.tsx - relationship state, tier,
list depth, activity presence filters and sorts.
8: apps/web-next/src/views/AccountProfile.tsx - per-person lists,
tier, last exchange, search count.
9: apps/web-next/src/views/Insights.tsx - strength tiers with
populations, stale-tie and re-engagement sets.
Footnotes
- dunbar
- connections
- index
- intel
- stale
- granovetter
- accounts
- profile
- insights