The Long Tail Isn't Winner-Take-All
Every directory story assumes a power law: a handful of winners take almost everything. Ours does not look like that — and the difference matters if you are the one publishing.
Across 765 skills and their accumulated download counts:
| Share of skills (ranked by downloads) | Share of all downloads |
|---|---|
| Top 1% | 5.6% |
| Top 5% | 20.2% |
| Top 10% | 30.7% |
| Top 20% | 48.1% |
| Top 50% | 78.8% |
For comparison, app-store distributions routinely have the top 1% taking 80% or more of downloads. Ours takes six percent.
19.9% of skills — 152 of them — have never been downloaded once.
Why the curve is flat
Three reasons, in order of how confident I am.
One: the directory is young and the traffic is thin. Total downloads across the whole site are still in the low thousands. At that scale, a power law has not had time to form — there is not enough traffic for rank-order effects to compound.
Two: much of the downloading is automated and broad. A bot that sweeps a category downloads many skills at once, which flattens the curve by pushing counts up across the board rather than concentrating them. A human choosing one skill concentrates; a scraper downloading forty does not.
Three: skills are cheap to try. Unlike an app, installing a skill costs a file copy. There is no purchase decision to concentrate demand. People can and do install several adjacent skills, which spreads the counts.
What it means for a new skill
The usual advice in a winner-take-all market is "do not bother." That advice does not apply here, for a specific reason: the top of the distribution is not defended by quality, it is defended by being listed.
The 18 skills at 9.5 and above average 32 downloads. The 165 skills between 9.0 and 9.5 average 8. That gap is large in relative terms, but in absolute terms it is 24 downloads — not a moat. A new skill that gets featured in one newsletter, or ranks for one long-tail query, can clear it.
The path is not "be better than the best skill." It is "get on one more list than the other 590."
The 152 that nobody downloaded
They are not the worst skills. Their average score is 9.03, higher than the 8.52 average of skills that have been downloaded at all. That sounds paradoxical until you remember what the download number actually measures: exposure, not quality.
A skill can be excellent and invisible. The distribution is flat because most of the directory is invisible, not because most of it is bad.
The caveat that applies to every number here
Downloads include automated traffic, and we have documented coordinated sweeps that download the same set of skills repeatedly. That inflates absolute counts and probably flattens the tail in ways we cannot fully separate from real usage.
What survives the caveat is the shape: there is no locked-up head. Whatever is happening at the top, it is not a winner-take-all equilibrium — not yet.