You can see the whitepaper here:
Weekly Consensus Meetings
The idea of a weekly meeting is similar to the standup meetings held by many companies.
All employees are used to attending weekly meetings and expect to get paid for doing so. It
is no different with members of a ƒractal.
The purpose of the weekly meeting is to build consensus on the rank-order value of each
individual's contribution to the ƒractal’s cause. Rank-order means sorting from greatest to
least. 4 out of 6 or 3 out of 5 must agree or no one earns anything.
The highest ranked individual from each of the groups is sent to the next round where the
process is repeated in a ƒractal manner (aka ƒractally). This process is repeated up to 5 times
or until there are less than 6 people in a round.
Schedule
All groups in a round meet at the same time. This prevents one person from having multiple
accounts and attending multiple meetings. The community governance process determines
the schedule for the meetings.
83% of the participants are only required to attend for 1 hour. 13% are required to attend for
hours and 2.3% are required for 3 hours. There is a 15 minute break between 1 hour sessions.
For really large ƒractals it may take 4 or 5 hours for the top 1% of most valuable members,but these members are also earning significantly more for their time. If you use a corporate
analogy, higher level management spends more time in meetings to reach consensus than
lower level employees.
Checkin
To operate efficiently and prevent empty groups from “no shows”, everyone must check-in
during the 10 minutes before the start of the meeting. This involves opening the ƒractally
application and visiting the meeting page. Everything from here is automatic.
Only those who check-in are assigned to groups. With the check-in process, there should be
very few people who are unable to join their assigned meeting and very few groups that
have 5 members instead of 6.
At the time of check-in a secret hash is submitted to the blockchain. After the check-in
window closes every client has 2 minutes to reveal their secret. Only those who reveal their
secret will participate in the meetings. The random groupings will be assigned based upon
the collective hash of all revealed secrets. Each person is thereby given the power to
change a single bit of randomness by waiting for “everyone else to submit” and then
deciding whether to reveal or opt out. For all practical purposes this ensures mathematically
provable honest shuffling of the groups.
Introductions
If a meeting starts at 6:00 PM, then everyone is expected to join the meeting by 6:05 PM.
These first 5 minutes are reserved for working out any connectivity, video, and audio issues.
People can also introduce themselves while waiting for everyone to join. Anyone who fails to
join the video call within 5 minutes should automatically be the lowest rank in the resulting
consensus. If more than two people are late, then the latest member should rank lowest.
This will be subjectively enforced by each group and any group that fails to enforce this
promptness rule may be held accountable by the governing Council. Established
communities should adopt a preamble (purpose) & an operating framework (rules such as
promptness rule) as part of their initial construction. The recorded video call should provide
all the objective evidence of tardiness required. The user interface will also enforce these
rules which should make it exceedingly difficult for a random group of 4 people to utilize an
alternative method of reporting a different consensus.
Presenting Work & Reaching Consensus
In each meeting, people are given a 5 minute time slot to present their contributions. After
all presentations are made, the group has 25-30 minutes to reach a consensus on how to
rank each member from highest to lowest. These rules are subjective and enforced via the
peer review part of the process.
Individuals can advocate for their individual contributions as well as the contributions of the
Team they have joined because the Team will receive 50% of the Respect earned by the
individual.
The group can use any process they like to reach consensus so long as everyone agrees by
the end of the 1 hour window. ƒractally intentionally avoided implementing a voting and
tallying system because all such systems encourage people to “vote strategically” instead of
honestly. Instead the meeting should be a back and forth discussion and negotiation.
The lack of a “voting” system means that people are forced to build trust that everyone is in
agreement so that they can accurately report their opinion on the consensus. Building trust
is a vital part of strong communities and identifying those who violate trust is critical to
securing the integrity of a community.
Proof of Consensus - Two Phase Commitment
Once consensus is reached, all members report to the blockchain with a two-phase commit.
First they submit a salted hash of the consensus opinion; then, after everyone has
committed, everyone reveals the consensus rank order. This process will be automated by
the user interface.
The purpose of the two-phase commitment is to prove that consensus was actually
reached. Anyone who wasn’t part of the video call and any automated systems would not
know how to report in. Furthermore, this is not a “vote” on what each individual thinks the
order “should be”, but rather, the mutual reporting of the shared consensus. At least 4 of 6 or
3 of 5 must report and agree. Anyone who doesn’t report a consensus aligned with the
majority does not earn any Respect. In practice, this means that all people will report the
same order.
Peer Review
At the end of each call, members can rate their experience with the other members.
Members who monopolize the call by talking too much, or are combative, rude, etc. should
receive a low rating. Members who are friendly, polite, courteous, and pleasant to work with
should get a high ranking. Over time everyone will be rated by everyone else and these
ratings can serve as evidence for toxic individuals whom the community may want to
consider removing.
Meeting Times
The governing Council of each community (more on this later) has the power to set the day
and time of each round of the weekly meetings. Communities that want to prevent
members from dual membership should consider scheduling at the same time as the
competing community. ƒractals that encourage dual membership may wish to schedule
their meetings on different days and times.
ƒractally - The Next Generation of DAOs 21
Respect Distribution
The following table shows how new Respect is allocated based on consensus meetings.
Round 1
Member Contribution Rank Round 1 Respect Earned
Rank 1 - Least Contribution 2 R
Rank 2 3 R
Rank 3 5 R
Rank 4 8 R
Rank 5 13 R
Rank 6 - Greatest Contribution 21 R
The contributors of the first round rank 1-6. Everyone at rank 6 of round one participates in
the second round and 5 of the 6 participants in that second round get promoted to higher
ranks (7-11). Because membership is not a multiple of 6, some groups will have only 5
members. In that case, the ranking is 2 through 6 and no one receives a rank of 1 from this
group.
This distribution pattern follows the Fibonacci sequence which is commonly found in nature.
Under this sequence, 16% of the participants in the first cycle earn about 40% of the
compensation. This is a softer form of the 80/20 Pareto principle.
Round 2
After the first round, those members who rank highest are then randomly grouped into a
second round. In this round, the Fibonacci sequence continues.
Respect Earned
Member Contribution Rank - Round 2 Current Round Prior Round (1) Net Increase
Rank 6 - Least Contribution 21 R 21 R 0 R
Rank 7 34 R 21 R 13 R
Rank 8 55 R 21 R 34 R
Rank 9 89 R 21 R 68 R
Rank 10 144 R 21 R 123 R
Rank 11 - Greatest Contribution 233 R 21 R 212 R
ƒractally - The Next Generation of DAOs 22
Note these are cumulative rewards meaning the net increase in the second round is 0 (21-21)
for the least and 212 (233-21) for the greatest. The goal is to rank-sort all contributors and to
have their reward grow according to Fibonacci.
Round 3
Respect Earned
Member Contribution Rank - Round 3 Current Round Prior Round (2) Net Increase
Rank 6 - Least Contribution 233 R 212 R 21 R
Rank 7 377 R 212 R 165 R
Rank 8 610 R 212 R 398 R
Rank 9 987 R 212 R 775 R
Rank 10 1597 R 212 R 1385 R
Rank 11 - Greatest Contribution 2584 R 212 R 2372 R
This process will continue until there are fewer than 6 people or up to 5 rounds.
In the event any group fails to reach a consensus, they receive Respect based upon the
consensus of the last group in which they were able to reach consensus. In other words,
instead of 5 of the 6 people being promoted from Rank 11 to Ranks 12-16, everyone would
remain at Rank 11. Consensus failure in the first round results in a Rank of 0 and no Respect
earned.
Anyone who submits a consensus report that disagrees with the majority will also receive a
Rank of 0 irrespective of the group’s actual consensus.
ƒractally - The Next Generation of DAOs 23
Ultimatum Game
There is a well-known economic experiment known as the Ultimatum Game
5
. In this
experiment, one person is given $100 and asked to share some part of it with another player.
The other player then has the option to accept or reject the offer. If the offer is rejected then
neither player gets to keep the money. If it is accepted then both players get to keep it. In
numerous experiments, it was found that offers below $30 were routinely rejected even
though $30 is better than nothing. The actual results varied from culture to culture which
indicates that ƒractals will likely work best when composed of people from a similar culture.
When asking a group of people to reach a consensus or get nothing, we face a similar
problem. If you put six people in a room and all of them believe they have made equal
contributions to a community, then they are likely to reject any consensus where they get
significantly less than someone else. By following the Fibonacci sequence, any two
consecutively ranked contributors agree to a 38/62 split in the award and are therefore
likely to reach consensus even if they both feel a 50/50 split would be fairer. The more
stubborn individual would likely get 62%. Furthermore, the group winner receives about 40%
while the rest, the runner-ups, collectively get 60%. This means that if you can view the
runner-ups as one logical person agreeing to the proposed split and the winner as being
asked to accept or reject, the winner would still accept even if no one else did anything to
deserve their compensation.
In a hypothetical situation where all members contributed an objectively equal amount, they
would have to decide between everyone getting nothing or allowing others to get more.
Chances are they would find some basis, perhaps even random chance, to reach a
consensus on the order rather than everyone getting nothing. Over multiple iterations of this
process, the community should correct for any past imbalances.
Any individual who is consistently part of groups that fail to reach consensus is automatically
removed from the community. This creates a bias toward building a consensus instead of
rejecting an unfair outcome. Those who are persistently stubborn would be naturally filtered
out to prevent them from griefing others. The proposed threshold for automatic removal is
failing to reach consensus 5 out of 10 consecutive weeks. Over a 20 week average, an
individual needs to reach a consensus 67% of the time to remain in the governing
community.
When you consider mimicking nature with Fibonacci, applying the experimental results from
the Ultimatum Game, and the Pareto Principle rule of thumb, we discover three metrics in
alignment which is a good indication the system may be balanced and aligned with human
nature.
5 Ultimatum Game, Wikipedia
ƒractally - The Next Generation of DAOs 24
Party Ultimatum
One of the potential outcomes of this game is that approximately 50% of the population
attempts to form a political party. This party attempts to demand that their members rank in
the highest 3 spots of every group of 6 and everyone else gets the bottom 3 ranks. To back
this demand, they threaten to block consensus and force everyone to get nothing. In this
game, the stubborn half takes 81%, and the passive half takes 19% of the rewards. Based
upon experiments with the Ultimatum Game experiment, the non-colluding parties are likely
to reject the efforts of the colluding parties outright. That said, the colluding parties can
bribe the first defector with the highest of the bottom 3 ranks.
Because 67% are required to agree to make a consensus and it takes 50% to block a
consensus. A two-party system would end up in a deadlock.
The reality is that any political party attempting to collude and split the bounty would still
require an internal party governance system. This governance system is necessarily less
democratic than the ƒractally process and would depend upon extreme party loyalty over
loyalty to the ƒractal. If such a party could create a better governance process then it might
just be beneficial for the entire community to adopt it. In any event, we speculate that the
incentive to form a minority party is relatively small relative to the gain party members might
expect from the attempted collusion.
Group Size
Eden
6 had previously used groups of 5 people. ƒractally has intentionally changed the target
group size from 5 to 6 in order to prevent a 40% collusive party from holding a 60% majority
hostage. Requiring 5 out of 7 to reach consensus would allow 42% to hold 58% hostage.
Going to 4 out of 7 would only require a 57% for consensus, which is significantly below the
byzantine fault tolerance
7
(BFT) level of 67%. The next closest group size that maximizes BFT
consensus and minimizes minority ultimatum is 6 of 9. The larger a group gets, the less
efficient a conversation becomes. A few people will tend to dominate the conversation in
large groups.
In a paper titled “Group Discussion as Interactive Dialogue or as Serial Monologue: The
Influence of Group Size
8
”, Nicolas Fay, Simon Garrod, and Jean Carletta studied the impact of
group size on communication patterns. They identify two kinds of communication: dialog
and monolog. In a dialog, members of a group discussion are likely to be most influenced by
8 Group Discussion as Interactive Dialogue or as Serial Monologue: The Influence of Group Size -
Nicolas Fay, Simon Garrod, Jean Carletta, 2000
7 While BFT is not relevant in the traditional sense, if you assume there exists a 33% “bad” minority that
is attempting to co-opt the governance process and a 66% majority which is aligned with the
intention, then it still requires 50% of the “good guys” to agree.
6 Eden - a blockchain community that utilizes a fractal governance process
ƒractally - The Next Generation of DAOs 25
those with whom they interact, whereas in a monolog they are likely to be most influenced
by the dominant speaker.
This paper indicates that increasing group sizes beyond 5 tends to increase the likelihood of
a monolog style discussion. Therefore, to maximize the influence of individual group
members, group sizes of 5-6 people are ideal.
While we are designing based upon theoretically equal groups, the reality is that not all
groups can have exactly 6 members (pigeon hole principle). This means that in any given
round up to 5 groups may only have 5 members depending upon how many check-in. It is
possible for more than 5 groups to have 5 members and some groups to have less than 5 if
some people check in but fail to show up in the meeting. When factoring in no-shows and
remainders, the target group size needs to consider the consequences of two group sizes at
the same time. So the options are 4-5, 5-6, 6-7, 7-8, or 8-9. Of these options, having some
groups of 5 is better than having some groups of 4. With a group of 4, 3 of 4 must reach
consensus (75%) and 2 of 4 (50%) can block consensus. If all groups could be groups of 4
then it would be a stronger consensus and easier dialog than if all groups could be groups
of 6; however, in this case, some groups would end up being 3.
After much consideration, a target group size of 6 with some groups of 5 was found to be
the best compromise. The rule for consensus within each group is either 4 of 6 or 3 of 5. This
means that it always takes at least 50% of the group to abort a consensus and at least 60%
of a group to reach a consensus, with the vast majority of the time requiring 67% for
consensus.
The pigeonhole principle states that if n items are put into m containers, with n > m, then at least one container must contain more than one item.