I've just posted the following ramble to the Chandler design mailing list. Reposted here as part of my recurring rambles about personal information management (and activity to which Outlook and its kin are an insult).
I've just come across clevercactus, which is interesting. It's a seems a bit flaky for a beta2, and the fonts are unusably small and apparently unchangeable on my 1400x1050 screen, but there are a few good ideas there. In particular the primacy of task over data type: it uses spaces for tasks (projects, whatever - things that you spend time on), and any data type can reside in any of those spaces, so you can keep (incoming and outgoing) email, RSS gathered news items and my own weblog posts, notes, contacts, where you use them.
Personally I think this still isn't quite right, although it's a sight better than the Outlook approach. InfoCentral, as has been mentioned before, got it as close to right as I've seen. It's all information, and it's all related. InfoCentral had a 2-pane view, if I remember rightly. The left was a tree pane, but it was a tree-shaped projection of a full graph data model. If you associated a person with a project, expanding the project in the tree would show the person (and all the other associated bits and bobs). Expanding the person would then show the project again. Not perfect, but a bit more space efficient than big graph diagrams, and I found it quite effective.
Even at the address book level, IC is the only tool I've seen which didn't leave me frustrated that the social or work arrangements of half of my contacts just didn't fit in the imposed model of what a person should do. One employer, one department. Several addresses, ok (not if you want to sync with a Palm, but that's another story), but with one-word labels. And if I have several contacts at the same company, and want to keep an entry for friends who are partners or just share a house, I have to type in the same information N times. Addresses belong to buildings, not people. People work at or live at buildings. People (may) work for organisations, and may work for several.
Another thing a decent data model should do (and another thing IC did) is allow you to declare a relationship between two "things" to have come to an end, without having to delete that information. When a contact moves to company Y, I don't want to delete the information that he used to work for company X. I want to add an end date the the connection, and create a new connection with company Y with a start date. Those dates of course were just one type of metadata which could be attached to associations.
So clevercactus, then. If clevercactus was a bit more stable, I'd use it, because it's better than anything else I've seen recently. I'll keep watching it.What it gets wrong is that it still imposes too concrete a data model, and loses some power in doing so. Contacts are _in_ spaces. I can't say, "here's a project, and this person is working on it, and this person is interested in it". I can't then define a filter which returns all people who are (working on | interested in) the project, give the filter a name, and use that name as a mailing list name. I want to be able to reify that dynamically generated group as well; there are interesting things to be said about it, like, at a really simple level, what messages I've sent to it's members.
From what I've seen of discussions and design plans for Chandler, I'm excited by the plans for the data store. No PIM software will meet my requirements which isn't built on a single coherent graph-based data model. I don't know enough about RDF (it's my reading list) but it's probably a reasonable choice and it has a fair amount of traction. I like topic maps myself, but that's not the point. The point is that *everything* in the system must be expressed -- whatever it looks like in any front end "parcel" -- in terms of that data model.
And talking of parcels, that's the aspect of what I've seen which worries me more. I may be wrong, but it looks alarmingly as if information is going to be partitioned based on it's type. The computer my care about it's type, I don't want to. I want all information that's related to what I'm look at now to be easily accessible, without big mode shifts. I'd really like my InfoCentral tree view back.
Ideally a parcel system would allow developers and power users (them being the same thing; the beauty of open source and agile languages) to easily build custom views of *subgraphs* from the data store. So for example a traditional address file view would pull in company and address info onto a typical view. Actions (such as establish a vid. conf. connection) could be encoded in the data model schema, and plonked pretty automatically in a right click menu at the level of the field or the whole panel.
Comments