I've banged on about context before and what looking at the distribution of objects that are by themselves inconsequential can tell us. This is nicely brought together in this month's Internet Archaeology where by looking at the datasets of finds distribution it was possible to make sensible deductions about the spread of anglo saxon settlement in England.
Now most of these finds were 'informal' finds made by metal detectorists, ie finds made by people who go out looking for things on a weekend. And if they find an anglo saxon penny, say, it's really cool. And if all that happens is that the penny goes into a finds box in someone's house that's the end of it. Interesting, even fascinating, but useless.
The value in the find is recording the find in a database. That way we know that a coin of a certain type was found at a certain location, ie context .If more, similar coins are found in roughly the same area, it suggests that something important was happening in terms of a cash economy - remember a silver penny was worth something like $50 - where large amounts of cash were being handled.
(In fact let's just say that a silver penny was worth looking for if you dropped it - like the man in rural Morocco I tipped 5DH - something like a dollar - for helping me. He said thank you, tossed it in the air and promptly dropped it in a pile of rocks. To me it was a dollar and if I couldn't find it easily not worth looking for - to him is was 5DH and extra bread for his family and so he set about fossicking enthusiastically to find it.)
So context and aggregation of data. Of course the datasets need to be preserved and publicly accessible to allow them to be cross referenced - meaning we can ask questions like 'do we find pennies on trade routes?', 'do we find pennies in locations where we find wine jars?' and so on.
And from a digital preservation point of view, the power of Julian Richards' Internet Arachaeology paper is showing what significant synthetic research can be carried out using publicly accessible but properly archived data sets - basically the power of dataset reuse.
And that is why we need to preserve datasets and make them publicly accessible - elsewise they're just a pile of spinning 1's and 0's ...