ciml - 2:09 pm on Mar 14, 2005 (gmt 0)

caveman:
> Now, the more you look at all the theories about kw lists that might be targeted by G for sandboxing, the less sense it makes.

There, we can agree.

> My answer: There is no list. There is no sandbox. There is only the algo and its associated filters.

The absence of a list doesn't by itself remove the "sandbox" idea.

People have found a long 'lag' from when a new domain receives its links, to those links counting. And yet, a similar page added to an established domain seems not to suffer from this 'lag'.

Many people have assumed that sandbox is a penalty, but I do not believe that we know this for sure.

The main non-penalty idea promoted to explain sandbox was that Google ran out of room at 2^32 pages, roughly 4.3 billion. This seems quite strange to me, as Google ought to have the in-house ability to deal with large numbers if they wanted. If I recall correctly, this idea had also been denied specifically by at least one Google employee (more, I think).

Another idea has been that Google are using some advanced kind of link analysis that takes a long time to calculate. We've read works by Jon Kleinberg, and by Krishna Bharat and George Mihaila. The aspect of this idea that I tend to dislike, is that Google would assign link 'theme' by domain and not by page. This would seem odd, although a number of people believe that pages on a well linked domain get an advantage, completely independent from PageRank. It would be fun to see if those two ideas could be combined.

caveman, shri, my mulling is simply that Liane's aside seems quite intriguing.

What if the Mr Véronis' idea of an index of two parts, was in part based on domains and took a long time to calculate?

I'm not pushing that anyone should believe this, nor even that the idea is at present the basis of a theory. I believe that this, like other ideas deserved some inspection.

steveb and shri both have examples of domains, seemingly sandboxed for searches that contain no words returning close to 80 million. That makes the idea seem less likely.