John, The reason for viewing an aggregate count rather than individual methods is to have a much more sensitive way of detecting a new topic. Sampling will only give us a good representative sample above the threshhold, once it's already quite popular. We'd like to be able to follow a topic from the start, even if it never makes it big, based on other criteria.
Rather than a trends feed, I suggest a statistics feed would be more general. Consider something like: http://search.twitter.com/stats/daily.json?target="foo bar xj17" which would give back the daily statistics on the frequency of each token. Then, we could follow up on individual ones of interest with a bigram request. With an api based on post, we could possibly send a large number of such requests in one batch. http://search.twitter.com/stats/bigram.json?target="foo bar xj17" (search for all messages containing any two of the listed targets) I don't know how you organize your data, but obviously, since you allow search by token, you have the infrastructure to make this easy to do.