How do CDB ratings work?

[CalebJ]

The way packages seem to be rated is non-descript and in some cases, rather strange.

For example, SimplePixel is rated at the #9 texture pack (~900 downloads and +3/-0 upvotes), whereas the Wilhelmines Texture Pack (~5000 downloads and +3/-0 upvotes) is rated lower, at #11, and Retro Plus is at #12 while having ~3500 downloads and +3/-0 upvotes.

The Games section is even weirder. There are all sorts of discrepancies, but I will focus on one: Glitch, with currently ~350 downloads and +11/-0 upvotes, is rated higher than Blockbomber, which has >6000 downloads and +13/-2 upvotes. If this is heavily biased by the upvotes/downvotes, then how is it that MeseCraft is rated higher than Glitch, which has >10000 downloads but only +4/-0 upvotes?

For what it's worth, I think Glitch is one of the most polished games I've seen, so I have no problems with that, but the rating doesn't make sense!

In the Mods section, the Automobiles Pack (~24000 downloads, +6/-0 upvotes) is rated higher than Animalia, which has ~47k downloads and +20/-2 upvotes!

So ... how are these ratings working? Or do they work at all?

[rubenwardy]

It is documented here: https://content.minetest.net/help/top_packages/

The page links to a REST API that will let you see where each package's score comes from

[CalebJ]

Thanks rubenwardy! The Frecency metric certainly explains some things. Maybe it shouldn't be weighted so much on this.

It felt like there was some logic regarding 'helpful' and 'unhelpful' reviews, I was disappointed to see that there is no such thing! This means that a negative review on a package could be downvoted 10 times but still affect the package's rating in the exact same way, and a positive review that is upvoted 10 times does the same thing as an upvote that has 5 downvotes!

Looking forward to discussion on the matter- It seems like there is a better package rating system to be found!

[Festus1965]

I read this and am near to vomit:

This is called a Frecency heuristic, a measure which combines both frequency and recency.

"Unique download" is counted per IP per package. Downloading an update won't increase the download count if it's already been downloaded from that IP.

Read it and know:
* reset Router, get new IP, download again = counts
* never us an update, always use the full set
to upvote a mod
* and just opposite to downgrade it

A download does absolute NOT tell anything about good or bad, just imagine a admin has to rebuild the hole set of his mods as of some failure don't find, and redu this for days ... funny he ?

The easiest rating is just every week go through the server-list and check most ACTIVE mods - that IS the reality.
And soon some useless server will come up, ONLY supporting own or favorite mods as of this als0 ... that is human.

But ... that is minetest - why easy, when they can do wired things ...

but as of the problems with TenPlus1 mods in ContentDB NOW I also understand why minetest leader don't do an own minetest git page and own server ..., same issue as had to refuse some mods as of license violation
Lawyers ARE expensive ... I know that !