Slashdot, and more Bram

[ciphergoth]

TrustFlow for LiveJournal has so for had nearly 70,000 different users, and over 0,6M hits. Users curemytragedy and seph were so curious to know their rankings that they reloaded the page, respectively, 595 and 817 times.

Ask Slashdot | Distributed Trust Metrics asks whether there could be a trust metric that works across websites. The question is vague and I don

Comments

[naturalborn]

The rate at which a bucket is filled is based on the upload capability of everyone who certs it, which is generally less than 1.

You can't exactly use condorcet because ratings aren't ranked. The latest proposal essentially does approval voting with proportional representation.

Source code is coming soon.

[ciphergoth]

I don't follow. Surely if we route all the water direct from the trust root to the non-trusted person, with none going to anyone else, then their fill rate is 1?

You can exactly use Condorcet the way I propose it. Supposing the candidates are A B C D, and two of them are to be chosen. Then you rank all possible results on your ballot paper: AB AC AD BC BD CD in whatever order you like. Or you assign 8 to A, 6 to B, 3 to C and 0 to D, and the following ranking is inferred:

AB (score 14)
AC (score 11)
BC (score 9)
AD (score 8)
BD (score 6)
CD (score 3)

Then you run Condorcet, once, over these six "virtual candidates" to get the winning team.

Of course, if you're trying to choose ten winners from twenty candidates, you have 670 billion "virtual candidates" and the run time gets a bit bogged down.

[naturalborn]

Let us consider a simple example. The seed certifies three people, A, B, and C, who certify, {D,E}, {E} and {F} respectively. After A, B, and C are filled up, E can be filled at a rate of 2/3, so it can be filled at time 3/2, while D and F can be filled at time 3. So we earmark the first 2/3 time of outflow from A and B to filling E. After that, E gets all of B's outflow since B is full, but for simplicity I'll assume that E doesn't cert anyone. Once we know that E will be selected, we have to compare D and F. F can still be filled at time 3, but D has to wait until time 2/3 to get its outflow from A, so D can be filled at time 3 + 2/3, so F comes second, and D third.

Condorcet can be used on whole sets of candidates, but it's usually done on one candidate at a time until they're all ranked. This approach is very much like condorcet but with a modification to do proportional representation.

[ciphergoth]

So the direct ancestors of the non-trusted person direct all of their flow to them, but everyone else distributes it proportionately? Why is this better?

I know that the way I'm proposing to use Condorcet is unusual, but I think it's a better fit for the way Condorcet works than an unnatural marriage with STV, which is what this proposal seems to be.