Agreed - it should all be in CPAN. The only reasons I suggest building a new site are:

• to test out the idea without bringing CPAN down
• easier to write new code with full access rather than having limited access to the code running CPAN (and I have no idea what state that is in)
• can experiment with ideas without annoying CPAN users with a changing interface

If the site proved it's worth, I would expect it to be integrated into CPAN, or at least for CPAN to provide a link to the module's page on this site.

The problem is the only definitive information about a module is it's POD (and code). Unless you try it, or you're very good at reading code, the only way to select a module is by what it's POD claims to do. Those claims may be incorrect, poorly communicated, or mis-understood.

I wouldn't even attempt to extract this information manually. I'm thinking more of wiki-style access (though maybe less free-form). I'm brain-dumping here, but how about a tree of "features", eg:

    Date/Time 
     |_ understand time zones
     |_ converts from RTF 1234 format
[download]

So the process would be:

• User finds module, wants to add Feature X
• Clicks : Edit feature list
• available features contains Feature X
  • Y: select Feature X
  • N: Add new feature
• Adds a rating 1-5 for that feature

Another user can:

• Select 5 modules to compare
• Site displays a table listing the union of all mentioned features of all the compared modules
• Modules missing a ranking for a feature displays [unknown] instead
• User has the ability to rank the [unknown]s

A feature-finder could be implemented from the same data, so you can add required or optional features from the feature tree, and the returned module list would be filtered (and ranked) to show those that support said features.

As I said, just a brain dump, but may have value. Any more ideas?

Clint