Peering Inside: Trafficking in meaning

Linden Lab is having a second stab at reworking the Second Life dwell (also known as traffic, but dwell is a more accurate name since traffic is not actually what is described) algorithm based on user-feedback. Last year, at Linden in-world office hours assorted feedback was gathered from users, and it appeared as though Linden Lab had settled on a new algorithm as a result.

This year however, Linden Lab is having another go at getting user-input for a new dwell algorithm, so we presume that something must have blocked progress on the previous attempt.

Users are generally divided, even on the simple matter of whether dwell calculations should stay the same, or be changed to something else. Of course, very few people actually know how it is being calculated now, or what the numbers mean. Unless you know what it is, and how it works, it seems a bit pointless to argue that it should be changed (or that it should remain the same).

Each account has a list of parcels visited between Midnight and Midnight. Only parcels which the avatar was on for at least five continuous minutes are added to this list. Any visits of length 4m59s or less are discarded and are not counted. If the avatar crosses a property line, that terminates a visit for that parcel.

Places in Second Life that utilize smaller parcels either because of administrative needs (streams, permissions, landing points) or simply because they are smaller are disadvantaged as avatars are more likely to cross a property line within the five-minute cutoff.

At the end of the day, a number of points are then allocated proportionally to the parcels on the account's list. We asked Linden Lab how many points an avatar actually received for allocation each day.

The answer came back, "one point" -- an answer that is demonstrably incorrect. If it is only one point, then the actual implementation of the dwell algorithm must be deeply and indeed grimly broken somewhere. While we've asked for clarification from Linden Lab on that figure, we don't expect to receive an answer for some weeks yet. A single avatar certainly distributes multiple points per day, as can be experimentally verified. Among those Second Life users who know the dwell system best, guesses range from approximately 100 to 255 points per avatar per day.

The dwell algorithm constitutes a one-way function in that the information inputs cannot be meaningfully derived or inferred from the resulting output. For any given traffic number there are combinations of times and visits that are functionally beyond reckoning that may yield that same result. Given two places with identical traffic numbers, one may be very busy, and the other significantly quieter.

This is because the current dwell algorithm depends more on what you do with the time that you are not at the parcel, than the amount of time that you are there. We think the one-way nature of the calculation was an intentional feature.

'The original idea was to reward builds of 'good' content,' says Linden Lab, when we asked about the original design goals, 'We first tried to detect "good" via Voting Stations -- green pillars placed in-world on which a user could click to indicate a parcel was "good". Only one vote per day per user per parcel was allowed. We would then tally up the votes nightly and distribute L$.'

Unfortunately, that got gamed to death rather quickly. 'Because users would gang up, fly around to all their friends' parcels and click the voting stations. Or they created vote farms - collections of 16m2 parcels each owned by a different person each with its own vote box, to make voting for friends easier. So we decided to replace it.' You'll still see those legacy voting stations around, but they no longer do anything other than clutter up the scenery.

'Dwell (subsequently renamed to 'traffic') was an attempt to limit how many "votes" a single user could provide in a day. The idea was to give each user one "aggregate vote" per day, and divide that contribution up among all parcels that user visited. Visits under 5 minutes were discarded as "just passing through".'

Dwell remained comparatively free of wholesale gaming for some while, but ultimately also fell prey to greed and falsity, as it was tied to three basic money-making systems, the ranking of Search results, regular L$ payments based on the magnitude of the traffic results, and period development bonuses for the highest ranking sites.

For those concerned with actual visit and attention metrics, of course, the current system is effectively useless. It tells you neither how much time was spent, nor how many people spent it. Two sites with the same figure may be (as we mentioned) hectically busy or eerily quiet, since the type of usage of the site factors in, as does the rest of the user's time in Second Life that day.

At the end of the day, the current dwell algorithm is a meaningful one -- but not one with human meaning.