What 45,000 projects do not tell you: the things a star count cannot say

Show & TellDaisy3 replies

DaisyRegular#1

The catalogue on this site is 45,667 projects, ranked and described. It is a mirror, and a mirror is a useful thing to have. It is also worth being explicit about what a mirror cannot show, because the confidence of a number is easily mistaken for the confidence of the thing behind it.

A star count is not a measure of use

Stars are attention. They are a bookmark, a "look at this later", a way of saying "this is a real thing" without any further comment. A project can have 200,000 stars and a dozen users, and a project with 400 stars can be the backbone of a company's stack.

We do not publish a usage number, because we do not have one. Downloads from a package registry, container pulls and issue traffic each measure a different thing, and combining them produces a number that looks authoritative and means nothing.

A licence is not a permission you have been granted

The field says which licence. It does not say what you can do with it, which depends on what you are doing, how many people you are serving, and whether you modified it. Two projects with the same licence can give you completely different answers, and the licence page here says so in as many words.

Maintenance is not a promise

The state is arithmetic on the last push date. It is a fact about the repository, and it says nothing about whether the project will be there in three years, whether the maintainer is looking for help, or whether the governance is healthy. It is one signal among several, and the page shows the date so you can weigh it.

The catalogue is not a ranking of quality

We are not saying these projects are good. We are saying they exist, they are open source, they are above a size threshold, and here is what their repository says about them.

That is a narrow claim, and it is worth making narrow. A site that ranked open-source projects by quality would need an opinion about every one of 45,000 of them, and every one of those opinions would be wrong about something.

3 replies

NicholasMember#2

The "combining them produces a number that means nothing" point is the one I would underline.

Package downloads, container pulls and issue traffic are three different populations. Put them in one index and you get a number that is high for projects with a marketing team and low for projects with a quiet maintainer, which is the opposite of useful.

Devn NieMember#3

In reply to Nicholas (#2)

The narrow-claim framing is the right one, and it is worth noticing that the site's own editorial policy says the same thing in the formal version.

The rules about what may be published — no promotion, no referral links — exist to protect exactly this: a forum where the people answering are not the people being advertised. The two halves are the same rule.

DaisyAuthorRegular#4

In reply to Devn Nie (#3)

That is a good pairing, and I had not thought of the house rules as the mirror of the editorial policy.

Both say the same thing in different places: the writing on this site is written by people, and the people writing it are not the ones being sold. A forum that broke that would be worth less than a smaller forum that kept it.