(Note: I was going through some old files earlier this week and found a couple of old posts that never made it into the live site. This is one of them. I've done a little bit of polishing around the edges, but this is as much a post for historical interest as is a reflection of the contemporary state of my thought.)

This post is a follow up to my not much organization post, and as part of my general reorganization, I've been toying with anything for emacs which is a tool, or set of tools, which provide search-based interaction with some tasks (opening files, finding files, accessing other information, etc.) in a real-time search-based paradigm. Mmmm buzzwords. Think of it as being like quicksilver or launchy, except for emacs. I've come to a conclusion, that I think is generalizable, but made particularly obvious by this particular problem space.

Search, as an interface to a corpus, is only more effective than other organizational methods when you don't know what the location of what your looking for is, or don't understand the organizational system that governs the collection where your object is located. When you do know where the needed object is, search may be more cumbersome.

This feels obvious, when put in this way, but is counter to contemporary practice. Take the Google search use case where you find websites that you already know exist. You'd be surprised at how many people find this site by searching for "tychoish" or "tycho garen blog." These are people who already know that the site exists and are probably people who have visited the site already. Google is forgiving in a way that typing an address into a search bar is not.

This works out alright in the end for websites: there's no organizing standard for mapping domain names to websites. This is mostly due to the fact that you don't, in the present practice, use the domain name system in the way that it was originally intended, in that the content of domain names are "brands" rather than a domain of systems and services described by the content of the domain. In the end this is not a huge problem since Google is around to help sort things out.

Similarly "desktop search" tools are helpful when you have a bunch of files scattered throughout file systems, with lots of hierarchy (directories and sub-directories). When you know where files are located, search less helpful. This is not to say that they're ineffective: you'll find what you're looking for, it'll just take longer.

I think this theory on the diminishing utility of search tool holds up, though I don't exactly know how to do the research to further the develop the idea in a more concrete direction. Having said that, I think the following questions are important.

  • Are there practical ways to organize our files, that don't require too much over-thinking before a collection grows unmanageable that make "resorting to search" less necessary?
  • Is (or might) building search tools for people who work with a given body of data (and therefore are familiar with the data, and are less likely to need search) different from building search for people who aren't familiar with a given corpus?

Onward and Upward!