Meta: Advogato's trust metric

Posted 23 Feb 2000 at 02:18 UTC by raph Share This

Reading through the diary entries today, I found that some people are quite unhappy with the way the trust metric operates, most vocally temas. I'm posting here to try to clear up a few misunderstandings, offer some suggestions, and possibly open a discussion.

First thing is that, in the move to the XCF, the trust metric got tightened by accident (I hadn't checked in the looser seed parameters into the CVS version of mod_virgule). I just fixed that.

Next, X-Virge misrepresents my intentions in this irc snip:

  • [18:26:11] <x-virge> raph started advogato as a sort of psychological experiment to see what would happen
  • [18:26:24] <x-virge> and this would prove that humans bring politics into things way too quickly
  • [18:26:35] <x-virge> even in a place where it really makes no difference

Yes, Advogato is an experiment, but (hopefully) one on how peers can define their own community in a way that keeps out trolls, spammers, kooks, and all the other types of antisocial behavior that have, unfortunately, come to define communication on the net. It is not an experiment to prove any bad things about human nature, especially politics. Free software has been pretty political from the beginning, and remains so in many ways (check out Debian if you don't believe me). This doesn't always have to be a bad thing, but that's just my personal opinion.

The trust metric (which is badly in need of a detailed writeup) is designed to keep out "attackers". As such, it works best when there is a rich cross-certification between groups of people. My belief is that there are many friendships and professional relationships that cross between projects and groups, so the basis for these cross-certifications is there. I've noticed, though, that people can be a bit shy about asking for certification. Maybe people should be less so. I'm a little reluctant to dictate what people here do, though - I want to see how the site works on a self-sustaining basis.

Finally, on the issue of the rankings. Please, don't take these too seriously. They are only a bit and a half of information, after all. That said, from what I know of the Jabber project, I personally would be reluctant to apply a Master label to its developers. Even though the project does appear to have lots of potential, there are after all quite a number of instant messaging apps out there. There's a good chance Jabber's unique use of XML and other properties will make it stand out among that category, and at that time I'd be totally comfortable with the Master ranking.

But that's just me, and I don't actually know much about Jabber. Hopefully, people who are more up to speed on the project will be using their own judgement, and end up with a ranking that's fair and accurate. That's the whole point of peer review.

I'm definitely aware of the fact that Advogato's trust metric is far from perfect. But I do think the ideas in it are one of the more promising ways to foster high quality discussion on the Internet. I'm hoping the experiment turns out well.


Appreciated comments, posted 23 Feb 2000 at 02:26 UTC by temas » (Master)

I appreciate this clarification and your comments. And I would like to apologize to all involved, I became hot headed and slightly askew with my normal views. I'm over it now. I hope this does not deter anyone's view of Jabber as it truly is an amazing project, with HUGE amounts of potential. I encourage you all to look at it and support it.

Thank you, posted 23 Feb 2000 at 02:31 UTC by julian » (Master)

Raph, I'd just like to thank you for clearing things up.

Unfortunately, my experience with politics has been mostly negative, but perhaps advogato will help show me a way in which having politics involved improves the situation.

Yes, I do agree that the jabber developers did take the rankings a bit too seriously, but I think a few more things were involved in their feelings at the time.

Anyway, hopefully things will be a bit better now. :)

Automated / Formula, posted 23 Feb 2000 at 02:57 UTC by eliot » (Journeyer)

Since we've got the attention of everyone, this is a good chance for me to ask a quick question which I have been trying to figure out...

I was under the impression that the trust metric automagically moved people up (or down) according to some system that will be published soon... Is this not how it works? Are there "admins" watching over to see who is ready to be moved around and "click the button" when appropriate?

Just kinda wondering how automated this was.. Some of the frustration was partly brought about because we (all of the Jabber developers involved in this) didn't think we would ever move beyond Observer simply because of a bias against our project. I was surprised because I was thinking that it was all automated and that someones (or a group of someones) bias against someone else would have no effect on the trust metrics -- even if the ones having bias had "admin" access. Don't take me wrong.. I'm not trying to accuse anyone of anything, I'm trying to understand exactly how the trust metrics work.

Re: Automated / Formula, posted 23 Feb 2000 at 03:02 UTC by julian » (Master)

It's fully automated. It was just that it was so tight that it required that one person be certified by several people of a higher level before that person's rank could increase.

This should sort things out., posted 23 Feb 2000 at 03:32 UTC by Radagast » (Journeyer)

I'm really happy that Raph wrote this. When I saw the diary entries for today, my first feeling was "Oh no, we're getting an anti-Advogato faction". Frankly, I think some people were taking things way too seriously, but it seems it's being sorted out now, so...

Random thoughts: It's probably not a good idea to go around ranking people as "Master", unless they're really widely known and widely respected. Alan, Linus, RMS, Miguel, Raph, Federico... Those are masters. If you work full-time on free software, and you put a lot of work into it, hey, you're a Journeyer. It's a great title to have. I'm very happy with mine. When that's said, I think the trust metric could do with a more sliding scale, but you've got to put the thresholds somewhere, so the system we have now works pretty well.

And in general, I'm not understanding the opposition towards judging other people. This is a meritocracy. If you do good stuff, you can be reasonably certain that someone else who does good stuff knows about you and your stuff, etc. That's how it works, and the trust metric seems to be doing a good job of modelling the real-life relationships.

Writeup of trust metric now up, posted 23 Feb 2000 at 06:35 UTC by raph » (Master)

You might be interested in a brief writeup of the trust metric used on this site. Enjoy!

subcultural divisions, posted 23 Feb 2000 at 06:43 UTC by rillian » (Master)

I was really happy to see this too, and looking forward to seeing better documentation. One thing that doesn't seem to be addressed in the current metric is any notion of subcultures. Perhaps this was intentional and/or we don't have enough cross-certification between groups yet, but I think this is what temas ran into.

I felt a similar isolation when I first came to the site (thank you lolo for ending my observer status!) My impression is that a lot of the people on this site are gnome folks, especially in the beginning, and so the trust valuations are biased toward members of that group. I don't know most of them. I'd heard of miguel and some of the gimp people, and that was about it. Not only was it hard for me to get certified, there weren't many others I could rate! Fortunately this is slowly improving.

These things are by their nature clustered. It would be nice if another group of developers, say the jabber folks, could set up their own patch of trust relations, and have that recognized independently. Is the current metric supposed to take into account that sort of meta-rating between clusters? I suppose I should look at the code to figure out what's actually going on. :)

In response to radagast's comments about rank inflation, that's not how I understood the levels as described in the instructions. For example:

A Master is the principal author or hard-working co-author of an "important" free software project, i.e. one that many people depend on, or one that stands out in quality. A Master has command of the tools and is an excellent programmer. Generally, a Master works equivalent to full time (or more) on free software. Ideally, a Master writes clearly about the work and its broader context, and serves as a mentor to others in the free software community.

Perhaps I'm just interpreting "important" differently, but I do think this reasonably covers people like dria and werner who I've certified as masters, though their fame doesn't (and at this point shouldn't) compare with people like Alan Cox. I still consider them the principle authors of important projects. And if we limit journeyers to those who work on free software full time, how many would we have? Just the lucky ones?

I do agree that we shouldn't be afraid of public 'rankings'. Of course it's fallacious to put much value on an index, particularly a one dimensional one, but I do believe a meritocracy is an improvement on typical power structures. The shyness here is no different from the 'dirty laudry' effects of holding open design discussions and open code development. Many of us were taught not to express publicly our honest opinion of others ("Don't ask someone how much money they have. If they make more than you, you'll feel shamed, and if it's less, they will.") but most of the same arguments for open development apply to social and political processes as well. To hide information is always to give power to those who would deal unfairly.

What the trust metric really measures, posted 23 Feb 2000 at 08:01 UTC by mjs » (Master)

I think the problem is that the Advogato trust metric claims to measure is community consensus about someone meeting the semi-objective definitions of the different certification levels. I guess the algorithm is meant to be designed to do this, while resisting attacks from parties no one trusts.

But in practice, it seems to measure the fact that at least 1 or 2 people who are highly rated like you enough to give you that level of rating. This certainly applied to me initially - two people whose names I didn't recognize rated me as "Journeyer" (a level which I think may have in fact been a slight exageration of my actual hacking contributions at the time). I've also been on the other side of this - I have single-handedly caused people to have a Journeyer or Apprentice rating by being the first to rate them, and in some cases maybe the broader community would not agree. Conversely, I've seen a lot of people not rated higher than Observer who had notes about interesting-sounding projects they were working on, whomI did not certify even to "Apprentice" because I didn't know them (and also, some of the projects seemed complex enough that certifying someone as "Apprentice" seemed more rude than not certifying at all; yet on the other hand since I knew I could make someone a "Journeyer" singlehandedly, I didn't want to do that without knowing about the project in detail).

Even though I know that's not the way the trust metric is supposed to work, the fact that it gives this appearance makes it have kind of a high school/IRC kind of atmosphere, where your "coolness" level is measured by whether some of the "cool" people accept you. Even Raph in his note above seems to imply that "Master" rating should be assigned based on fame.

So I have decidedly mixed feelings about it. It's a very interesting concept; it tries to measure directly the "peer repute" that Eric Raymond has written so much about. But it seems in practice to result in the sort of cliquish environment I have always hated elsewhere.

Correction, posted 23 Feb 2000 at 08:06 UTC by mjs » (Master)

That was Radagast, not Raph. Wish I could edit posts.

patch, posted 23 Feb 2000 at 09:35 UTC by mbp » (Master)

I'll leave aside reservations about whether pigeonholing is good, and instead suggest a change to the algorithm.

Perhaps people should say for themselves what level they think best describes them. Other people can decide if they think it's accurate or not. If they disagree, they can write to the person and explain why it should be higher or lower. If they agree, then they can `sign' the claim to indicate approval.

So, we have two variables for each person: the level they claim, and the credibility of their claim.

People who are rated too high will be there because they have an inflated opinion of themselves, not because other random people were trying to be nice.

As a refinement, perhaps you could register your disagreement with a signature, which would result in an amount proportional to your own credibility being subtracted from the strength of the signature.

I haven't thought through whether this has the property of being safe against many attackers; I think it could be made so.

Some thoughts, posted 23 Feb 2000 at 10:22 UTC by Skud » (Master)

I signed up last night, and came back today to find that I'd been certified at "Master" level. This seemed odd to me. Yes, I work full time in Open Source and have been involved in various projects for, well, ages. Yes, I do a lot of advocacy, writing, and mentoring. But as mbp points out in his diary, nothing I've done has been a "moby hack".

It seems to me that the problem in my case is that while I've been cutting code and contributing to projects for years, none of the Open Source coding I've done is of particularly ground shaking importance when taken in isolation.

I don't really like to make assumptions, but I think when dria certified me, she was probably not taking my code in isolation. Dria knows me through my documentation and other writing, and through my community involvement.

Now, Advogato's description of the levels says that a Master (and I won't even think about mentioning "Mistress" as an alternative title) should be the principal author of a major piece of software that many people depend on. It's biassed in favour of coders, and a Master-level documentor, advocate, teacher, project manager, or integrator cannot be recognised as such under the rules as they exist.

So people have bent the rules. Looking at the other Masters on the system, I can see:

  • A lead developer on a documentation project (no coding projects listed)
  • A contributor to GIMP (no other projects listed)
  • A couple of people who list no projects

My guess here is that some users are playing by the rules (which is fair enough) and others are interpreting the rules to allow for non-code contributions. Whether the latter is a reasonable thing to allow, I don't know. This is something that advogato's administrators and the advogato community need to think about.

My vote is to recognise all contributions to Open Source/Free Software. I'd tend towards loosening the rules, and letting the trust metric do what the trust metric does. Otherwise, the two behaviours will cause glitches in the way the trust network works, and we won't be able to clearly see whether the trust metric is working.

Popularity?, posted 23 Feb 2000 at 11:32 UTC by rakholh » (Journeyer)

I think I was the "spark" with this whole issue. I commented on irc.gimp.org on #gnome - advogato.org is being invaded by 'Jabber' people. I meant that in a 'nice' way (i.e. Jabber-folks disocvered advogato.org and all signed up).

Anyway, I think the problem is that temas and some other Jabber-folk are too 'biased' towards their project. They view their project as important and claim that other companies want to use it. That may be all nice and well, but the fact is that we haven't actually seen it deployed. Nobody I know except a few folks use Jabber. It is not a 'popular' project at this time, and is under heavy development. Don't get me wrong, it /might/ be a good project, it might to the InstantMessaging world as Apache was to the WebServer role. But that is not the case currently.

What people are objecting to is the fact that you Jabber folk are ranking each other. While you may /believe/ that you are objective - you really are not because you interact with these people daily and while /you/ may 'trust' them the advogato.org community may not because they have no clue on Jabber.

Some people probably take 'offense' at the fact that you're ranking each other on the leagues of 'Alan Cox' or 'Raph' or 'Miguel' - They are popular in the advogato.org community because they have /demonstrated/ to everyone their technical talent. Since Jabber is not yet popular, people can not judge you based on your technical talent (since they don't use Jabber and aren't impressed with it).

Like Raph said - don't take it personally. Its all a matter of perceptions :)

(eql "ability" "effort") => NIL, posted 23 Feb 2000 at 11:39 UTC by dan » (Master)

Skud writes:

My vote is to recognise all contributions to Open Source/Free Software. I'd tend towards loosening the rules, and letting the trust metric do what the trust metric does.

<aol>Me too!</aol>. Aside from that:

One feature of the certification system used is that it conflates effort and ability. You can't be a Master simply by being very good at what you do; you also have to do a lot of it. At the other end of the scale, if you would only rate Apprentice because you only have a few hours a week to spend on free software, you also get labelled as "still striving to acquire the skills "

One probable result is that many people are shy of certifying others at "Apprentice" level, because - well, because it sounds demeaning. I'm happy to acknowledge that I don't work on free software except at the weekends (and what I do then is also of limited general interest, granted) but I'd resent the implication in Apprentice that I don't know what I'm doing.

Simple fix: what say we deprecate "Apprentice" and introduce "Royalty" or "Omniscient being" or something at the top of the scale? We still have the majority of people at "Journeyer" as now, but there's an extra bit to differentiate between, say, Dria and Alan. I hope neither will take offence at my using them as an example.

Far-reaching change: introduce two orthogonal ratings for effort and godliness.

Re: ("ability" != "effort"), posted 23 Feb 2000 at 13:20 UTC by Raphael » (Master)

dan writes:

Far-reaching change: introduce two orthogonal ratings for effort and godliness.

Yes, that would allow one to say: "Mr so-and-so has great skills but has not contributed to any major software packages lately, so I will rate his ability as Master and his effort as Apprentice." That would probably remove some ambiguity in the current ratings. And that would hopefully encourage more people to rate others, because it would be easier to select the level for each criterion. Currently, it is a bit tricky to evaluate the level of someone else because you need to take a subjective mix of the two criteria.

The "ability" category could use the same level names as the ones used now. The "effort" category could consist of the following levels:

  • Prolific
  • Productive
  • Promising
  • Sleeping
(Note: I am sure that some native English speaker can find better words for that.) This could lead to something like a "Prolific Master" or a "Sleeping Journeyer" (and what about the "Promising Mistress"?).

In another comment, rillian mentions the subcultural divisions and writes:

My impression is that a lot of the people on this site are gnome folks, especially in the beginning, and so the trust valuations are biased toward members of that group. I don't know most of them. I'd heard of miguel and some of the gimp people, and that was about it. Not only was it hard for me to get certified, there weren't many others I could rate! Fortunately this is slowly improving.

The problem is that you are supposed to certify people you know and it is difficult to judge people that you do not know. This could be partially solved by encouraging newcomers to post more information about themselves, so that the people who are already certified could have a means to know more about them and maybe certify them (assuming that the sources of information can be trusted).

Another way to improve the current situation would be to allow tentative certifications or weighted certifications: when you certify someone else, you would have the opportunity to assign a weight to your certification, based on how well you know that person. That would basically allow you to explicitely reduce the weight of your certification in the certification graph. So for a person that you have met personally or with whom you are exchanging messages frequently, you would use the full weight. On the other hand, you could help a newcomer that has not been certified yet by giving him a tentative certification that has a lower weight.

There are many more things that could be done to improve this system. I wrote last week that I would post a new article about that soon, following my previous article on Slashdot moderation. Hmm... I will try to do that next week...

Mixture of concepts, posted 23 Feb 2000 at 19:56 UTC by graydon » (Master)

It seems to me that advogato's trust metric mixes a number of concepts.

First, it mixes how much I trust someone with how good a hacker I think they are, which are not synonymous. This has the unfortunate effect of granting master hackers I might admire at a distance an unnecessary and frankly unwanted level of significance in choosing who I get to hear from and who I trust. If I call someone a master, it is out of a judgement of their skills. They might be a jerk. More often than not, I don't even know if they are a jerk. I should not be implicitly saying they are not-a-jerk just because I happen to be using one of their libraries.

Further, it mixes the concepts of who I certify and who I trust the certifications of. In other words, I cannot grant trust to a person without approving of their ability to judge others trustworthyness. Lets say mike thinks I personally am trustworthy, but hopelessly naive and unable to tell good people from bad. He then has to choose whether to put me in his trust network (and infect it with all my stupid judgements) or leave me out of it, and subject me to a life of not-being-certified-as-trustworthy-by-mike. That's just not correct.

More troublingly, advogato's trust metric presents a trust network with its origins in 4 people who are not me, which is clearly incorrect. Trust flows from a person's own judgements, not from a central registry of what is right and wrong. This is the post modern era -- we have dialogues and metadialogues but not truth. While it is certainly interesting to be able to see who raph, alan, federico and miguel trust (and what their trust network reveals about everyone else), it really doesn't represent the trust relationships of everyone in the free software community "at a glance" any more than a headline on CNN represents "the news at a glance". It's one flavour of trust, from one source.

Advogato would benefit from the following modifications:

  • a logged in user has a trust web calculated on the fly with them at the root.
  • trust, righteous hackerdom, trust-to-introduce and trust-to-judge-others-hackerdom are separete certifications with separate network flows.
  • people can see one another's trust webs if they like, but are not by default logged into raph, miguel, federico, and alan's trust web.

Multiple trust webs would fragment the community, posted 23 Feb 2000 at 20:56 UTC by Raphael » (Master)

graydon writes:
a logged in user has a trust web calculated on the fly with them at the root.
[...] people can see one another's trust webs if they like, but are not by default logged into raph, miguel, federico, and alan's trust web.

Although this solution would be nice for some individuals, it has several drawbacks:

  • It would lead to a fragmentation of the community. We could end up with several disconnected trust webs, causing some people to read and post articles that are only visible to a small group of users and ignored by everybody else.
  • Advogato would have to accept and store all submitted articles even if they could only be read by their author (because even if the author has not been certified by anybody, he is at the root of his own trust web). Some authors might not even know that nobody is able to read their articles.

Last month, eivind posted a comment explaining the fragmentation problem. Here is an excerpt from the end of his comment:

It causes opinion reinforcement for the users - users see messages that agree with their opinions, represented by what they have moderated as 'good' before.
It causes opinion-based splits of the community to sub-communities - different parts of the community see completely different discussions.

Although it should be possible for the users to configure their "view" of the articles (using priorities, thresholds, etc.), Advogato should try to keep the trust web as connected as possible. Otherwise, a group of people who trust each other could use the site to discuss sheep herding while another group is exchanging cooking recipes and another is talking about politics. By having a limited number of people at the root of the global network of trust, Advogato ensures that the topics and ratings do not drift away from the original intent.

Ah well... The best solution is probably somewhere inbetween...

2 points, 2 refutations, posted 23 Feb 2000 at 22:11 UTC by graydon » (Master)

The system would fragment!

Incorrect. The number of connections between individuals remains the same regardless of the metrics assigned. I know bob, bob knows me, no amount of metric tweaking will change that fact. What does change is whose metrics are being poured into the network when solving for the trust flow to a given individual. In other words: when I examine bob's page, does it classify him as trustworthy because of judgements I've made, or judgements raph, miguel, federico and alan have made? I submit that the former makes much more sense.

Anyone can post anything!

Again, you are blending the issues. It is not necessarily the case that the system will accept postings from anyone. Anyone can post followups to slashdot, but not anyone can submit lead stories. Likewise here, there's no reason to believe that just because your trust metrics are separately maintained, that the owners of the site (who pay for its facilities somehow, perhaps through academic slavery :) will not reserve editorial control over the parts of the site which are justly determined to be scarce. Like any setting (IRC, netnews, mailing lists, web sites) there are sections which are hand-moderated, sections which are mechanically moderated, and sections which are free-for-alls. The right balance of these moderation strategies will be made irrespective of the trust metric choice.

I maintain that the only reason this issue isn't more blatantly clear is that most of us assigned master-level trust to some if not all of the 4 roots of the trust web when we joined. This coincidental action cannot be misconstrued as proof that the web assigns metrics in a logical manner.

From the back of the class...., posted 1 Mar 2000 at 07:45 UTC by kmself » (Journeyer)

For someone who's only studied lower math (and forgotten most of it) -- calculus, some diffy Qs, and a bit of numeric algebra -- is there a translation of Raph's very interesting white paper? I'm not familiar or comfortable with treatments using graphs, nodes, edges, supersinks, and the like. Much of the language of the paper is over my head.

FWIW, Google turns up a couple of scholarly treatments in the first page of results, including a Japanese paper with Java applet demos and another paper by Harvey Greenberg at University of Colorado, Denver.

While I think I follow some of his points, it's not clear to me in the source/sink discussion, what is being sourced or sunk -- is it traffic or trust? In the attack model, how are nodes assessed as "good" or "bad"? This appears to be an arbitrary or normative decision at some point. It would seem that "confused" nodes are those linking to both bad and { good or confused } nodes, though Raph's illustration includes one "compromised" node which links only to another compromised nodes and good nodes. I'm confused. I hope I'm not bad....

Again, in the event of an attack, how does the model translate to a defense strategy? What is the model buying that isn't available by other means -- single-point moderation, selective filtering, etc.?

What also of the instance of "good nodes gone bad"? Several well known attacks on Slashdot where performed by "karma whores" -- users who'd accumulated high karma scores which could be "spent down" on spam attacks against the board. How does the trust metric allow for detection and action against such behavior?

I believe there are several people reading who would appreciate a layman's explanation of this concept. TIA.

Some comments on the trust metric, posted 1 Mar 2000 at 16:05 UTC by ajt » (Master)

A cute abuse of Avocado's (you know you want to call it that, why fight it?) trust metric is nwv. It's cute because as well as being your average anonymous account , it uses a nice social hack that could, conceivably, manage to make the creator anonymous at some point. (There's not much point for a Journeyer, say, to certify some random Observer; but for a Journeyer to certify someone else who's already a Journeyer, heck, why not? It can't hurt, and it increases the connectivity of the trust graph, and hey, if he nominates me as a Journeyer, I'm that little bit safer, right?) When the dummy account has built up its web of supporters well enough, the original certifier can quietly disappear from the account's list of certifiers, and sooner or later do whatever it is you do when you've broken Avocado's security.

Of course, breaking Avocado's security is probably an exageration: Avocado is meant to stop large scale attacks with lots of fake accounts; it's not meant to stop every lamer from sneaking through the cracks.

Perhaps one problem with this is there's no negative reinforcement. If someone thinks you're being a twit, and gives you an Apprentice or an Observer rating, well, no harm done. Even if fifty people give you an Apprentice rating, it doesn't really matter, as long as you've still got one or two Journeyer ratings.

Failing any sort of negative reinforcement (which could just end up an excuse for angst, or revenge, or more certification wars, or whatever) there's not really any sort of ability to limit how much you trust someone. One way to do this would be to disregard the mathematical network flow model, and follow a more computery one by adding a TTL field to your flow, of some sort. That's probably painful and tacky, and hard to analyse though. Another option would be to make the capacity of an edge more variable: if I trust someone enough to think they're a Journeyer, but not enough to really want to let them make anyone else a Journeyer, I can make the capacity of their link just `1', which will presumably go straight to the sink. If other people decide they like this guy too, then he can use their flow to certify other people. (But what if they don't want their flow to be used that way? Oh well.)

In any event, the `capacity is proportional to distance from the source' seems a fairly arbitrary metric.

Another thing I wonder about is why only Master's should be able to deem other's Masters, and so on. (This is one issue with sourcing trust at "yourself", rather than alan and co --- if you're not a Master to start with, you can't set things up so that anyone else can be a Master. You could arbitrarily declare that everyone by default considers themselves a Master, but that's probably not true. Or you could just special cast it) A possibility here would be to connect everyone at the next lower level to the super-source, but giving them a very limited capacity. So that while Alan can make anyone a Master with a gentle nod (because his capacity is 1000 units, and it only takes 10 to make a Master, say), it takes significantly more Journeyers to do so (because their capacity for assigning Masterdom might be only 1 or 2 each). Furthermore, while Alan can make 100 people Masters, each Journeyer can only contribute to a couple of people being ranked as a Master.

Non-determinism would hurt pretty badly here, though, unfortunately, so your flow might happen to make a bunch of .9 Master's (which don't count) and no 1.0 Master's (which would count). I'm not sure how hard it would be to work with a network with some binary edges: ie, edges to the super-sink getting a flow of either 0 or 10 exactly. Tricky.

A nice feature, btw, would be to be able to see where your trust is coming from on your homepage. `You're a Journeyer: you got 5 units from foo, and 3 units from bar, and 6 units from baz, of which you're passing 6 down to quux, and 7 down to quuux'...

Hmmm. That's strange. My <p> tags got doubled up in the edit window of preview.

Bruce Sterling's "Distraction". Timed-weightings., posted 5 Apr 2000 at 04:20 UTC by lkcl » (Master)

For those people interested in trust metrics, which for some reason I am one, Bruce Sterling's book, "Distraction" is an interesting read.

It describes a nomadic, non-monetary society (outcast by the now-bankrupt U.S. government), who allocate responsibility based on a trust metric. lose trust, you lose responsibility, and have to earn it again if you want the responsibilities back.

It also has the very neat side-effect of making individuals replaceable (difficult to target the head if it's a hydra!) by the next-most-responsibly-trusted-person, automatically.

Thoughts on Trust Metrics:

i think it would be good to have the trust metrics measured on the way neurons work. namely:

1) an advocacy decays, or is given less weight, relative to more recent advocacies. this has the side-effect of allowing lots of little-weighted-advocacies to build up the same clout as one big-weighted-advocacy, and also reduces the effect of Past Offencies Against The System (forgive, but do not forget) and likewise for older contributions.

2) a negative-rating system with equal clout as the positive-ratings. Such names as, "idiot", "offender" and "outcast" spring to mind as possible equivalents to "Apprentice", "Journeyer" and "Master". a negative rating against an individual would bust them lower than observer, and their privileges restricted as a result, with difficult but not impossible means to reinstate their status (weighed in the balance, and found wanting :)

3) an advanced system of "group trust". a group of individuals (or a possibly recursive system, a group of groups!) get together and create a group, membership by invitation only. they decide to allocate trust-metrics to certain individuals to certify that those individuals are trusted inside *their* group to represent them and the combined trust-metric of their group. if the group's individual members decide that their own "trusted representatives" are being stupid, and bringing the group (and therefore the individual's) trust metric into disrepute, they can decide to *renege* on the trust-metric they gave to their representatives, or just simply leave the group.

the "group trust" system is a really interesting concept, and mirrors the way that societies work. there is always, within a group, an individual or set of individuals that are trusted to represent the group. the individual contributes to the group by their own trust-metrics, and if they leave, the group loses their trust rating. if the person was a valuable contributor to that group, the group might lose its trust-metric sufficiently for other members to *also* leave, forcing the collapse of the group and reforming of a more suitable one etc etc.

4) the creation of an internet protocol (a simple one! learn from microsoft's mistakes!) to provide distributed trust-metrics. for example, i would like to see one per project running *on* each project's site (e.g samba.org, i have to talk to the other samba team members about this). this to help alleviate any issues with 3) - a group itself might not want to trust another group's site to run their own trust metrics!

5) a programming api to allow systems such as samba-patches@samba.org to automatically generate (or help generate) trust-metrics for contributions to projects. samba, for example, has a simple version of trust metrics, already: number of cvs commits (by samba team members) and number of patches submitted (by non-samba-team members, who do not have cvs access to samba.org).

anyway. these are some of my thoughts over the last few days and nights. based on bruce sterling's ideas, i am so pleased to see an actual implementation of a real trust-metric system, i'd love to see it develop more. yes, i can help out with that :)

More than 8 years later, two links, posted 22 Oct 2008 at 11:33 UTC by chalst » (Master)

This article provoked some thoughts of mine about the possibility of allowing alternate centres for advogato. I tagged in on as postscript #2 to my diary entry from 2003 starting with Afghanistan, Iraq and a mea culpa on the Bush tax cut. Those were the days...

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!

X
Share this page