Here are the problems I observed within the first five minutes of using
the damned thing:

1.

Simple search IR algorithm BLOWS GOATS. It seems to take term frequency
in the entire index into far too much consideration when averaging out
whatever other hitters they have (term frequency in doc divided by
number of total words in doc, if they're smart).

And whatever IR they do gets blown away by simple search's insistence
that the term appearing in the title take precedence over all else.

2.

The natural language query, while pretty, needs an English lesson. Try
it yourself if you don't believe me. Larry F. & Co. would wince at what
it does to a natural language query for "leaves".


3.

Power search supplies most relevant STARTS-style operators, but is
missing one of the most important: exclusive or (XOR).

In addition, it tries too hard to do proximity searching-by-proxy, with
an automatic setting of "Medium" (on the frightfully useful L / M / H
scale -- duh) for the importance that the words appear in the exact
order you supplied them.

Excuse me?

If it's a phrase, order counts. Likelihood that anyone, power searchers
included, will EVER type two words close to one another that are not
part of a linear phrase: About ten percent. And the horndogs that type
"sex slut tits pamela anderson" in the simple search box deserve to find
nothing. Either acquaint yourself with power searching r take a cold
shower, boys.

4.

On the summary generator: it isn't. "Standard" summaries are the first n
characters of text after the first <p> tag. "Enhanced" summaries add
valuable insight into the page's content by supplying the first n+10
characters.


Brilliant engineering work. Must've taken hundreds of minutes of R&D to
implement.


5.

The site itself is ugly. As sin.

Black and Yellow? What is this, 1995?

And get those damned ads out of the main pane and into the bottom pane
where they belong (that is, where I can ignore them).

My $0.02.