When geocoding a large dataset, it's common for some of the results to not get exact address matches. Most address locators (including, presumably the one supplied by Esri that Socrata is currently using), use a cascading system that will first try to match based on address point (often rooftop/ most accurate), then by street address range (not as accurate as rooftop, but still very accurate), then zip, then city (least accurate).

I think it is important for people users of Socrata's geocoding system know the limitations of the geocoding engine, and that there results are not going to be always 100% accurate, and sometimes may not get geocoded at all (I've attached a screen-capture for an example).

Another possibility is that the type of locator used (e.g., address point, city center, etc.) and the math success (usually represented in %), could be added to the table, just as the actual lat/long are added to the location column.