[Data-modeling] the commons / non distinction is a farce

[Kirrily Robert]

On Jun 29, 2009, at 2:46 PM, Tom Morris wrote:

> On Mon, Jun 29, 2009 at 2:39 PM, Kirrily Robert<kirrily at metaweb.com>  
> wrote:
>
>> The email notification system that just came
>> live is a help as it allows us to contact schema admins more easily.
>
> I wouldn't count on this necessarily helping.  I was quite surprised
> recently to discover that I wasn't subscribed to updates for a domain
> that I created, so   I just created
> https://bugs.freebase.com/browse/FREEBASE-834 suggesting that admins
> be forced to listen to updates on their domains.

This is how it currently works, in fact.  You are not "subscribed" in  
the sense of actively following a domain you admin, but all admins of  
a domain (base or commons) do get notifications for any discussion  
posts on their domain's homepage.

>> Acre
>> code search is likely to be increasingly useful for seeing just how
>> many people are using a certain schema in their apps (it won't catch
>> non-Acre apps, but as the number of apps grows, it will at least give
>> us *some* idea.)
>
> If this is to distinguish things that need non-breaking refactorings
> vs those that don't, I think you should assume that every piece of the
> schema is being used and refactor it in a non-breaking fashion, if at
> all possible.  If it's to determine "popularity" of particular
> domains/types/properties, I'd suggest that doing it by query count
> would be a) more realistic in terms of popularity and b) fairer to
> non-Acre apps.

There are some things that simply cannot be refactored in a non- 
breaking way.  For instance, changing a property from unique to non- 
unique.  There's no way to do that while guaranteeing backward  
compatibility.

When that sort of thing comes up, we make a judgement call on just how  
much notice is required, based on how much the types in question  
appear to be used.  We look at both the number of links, how common  
saved views are against a type, and how common apps are against a  
type.  It's a bit of a black art, but we can judge all these things  
based on what's in Freebase, *except* for external (non-Acre)  
applications.  The help topic on "Freebase Commons admin guidelines"  
at http://www.freebase.com/view/guid/9202a8c04000641f800000000b75f213  
describes how we usually handle it.

Anyway -- there'll never be a way to tell exactly what people are  
doing with the API or data dumps, but Acre code search may help us  
guess at the popularity of a type.  For instance, if there are many  
Acre apps, then there are probably many non-Acre apps too.  If there  
are no Acre apps, then while there *might* be a bunch of non-Acre apps  
that we don't know about, the likelihood is less.

K.

-- 
Kirrily Robert
Freebase Community Director
kirrily at metaweb.com
http://freebase.com/