Rootconf talk

In November 2025 I made an elevator pitch for a talk at Rootconf about creating infrastructure for volunteer communities. This went really went in that I had some cool ideas that the pitchees also thought were cool, and so far it's gone poorly in that I've yet to finish incorporating their feedback to add back the technical depth required to do that topic justice. Mostly that's because the scope creep ate my brain for awhile. There are a couple of different directions I could take for technical depth here, and they don't all fit in one talk.

So let's regear and start from scratch.¹

There are a few central insights I want to capture.

A coalition is like a vehicle route.

A coalition is the game-theory concept for a collection of agents for whom working together is better than working alone.

Think of a coalition as a group of people, moving in the same direction, in a vehicle. this is the analogy I was taught coalitional theory with, and need to find the original source of this construction.

The coalition as a whole selects its path on the basis of some consensus value – but the concensus only needs to hold for current passengers, about the stops those passengers care about. People get on or off at the stops they were trying to get to. While at least one person on the vehicle wants it to go somewhere enough to pay for gas, the vehicle keeps running.
All passengers care about where they get on the vehicle (origin) and where it drops them off (destination).
Passengers are usually pursuing incentives of their own, which the vehicle is instrumental to.
the vehicle is costly to run, i.e. passengers have to agree on a payment model to keep it running. If it isn't profitable for them, as measured against their incentives, they won't get on.

You can see that with this level of generality, there are lots of potential kinds of solution. Let's look at two.

A vehicle might work like a bus, i.e. its passengers don't tend to share origins or destinations. Call this common-means coalitions.
OTOH it might work like a carpool, i.e. passengers share a common origin or destination. Call these common-origin and common-goals coalitions.

todo write out the formal construction of the vehicle route.

Orgs are forcing functions.

Volunteer orgs are common-goals coalitions computing a forcing function.

All coalitions can fail by inadequate profit, but common-goals coalitions will additionally fail by goals divergence. If a goal is only approximately understood, and over the course of computing it you find that your incentives don't select for it anymore, you might want to leave; but this may strand you somewhere you don't want to be, and you have a much higher chance of stopping the vehicle altogether.

Volunteers are a fundamentally different type of agent than employees are.

Employees are invested in the vehicle paying them in utils. The vehicle is a vehicle. It has cargo that has to get somewhere.
Volunteers are invested in the path of the vehicle. They want the cargo to get somewhere.
Compensation-seekers get on for-profit vehicles. Outcome-seekers get on volunteer vehicles.
Compensate volunteers in cargo space. Compensate employees in utils.

Analogously, volunteer vs for-profit orgs are different types of agent.

A volunteer organization is always a community.

You are always building a collection of people with shared goals and values, who must work in collaboration to define and operationalize their personal goals and values. This requires them to have high trust in each other and the group as a whole.
- However you define community, putting humans in a group with these properties will put them in community.

Path capture

Almost certainly, per Arrow's impossibility theorem, a smaller agent-unit has control over the truck's path. Not all truck passengers can be drivers.
All agents across all types of organization are potentially interested in path capture – i.e. controlling what the truck carries and where it goes. The resources which they have for doing this, and the internal incentive structure of the truck-coalition that they are navigating to do this, differ fundamentally.
In practice, the agents most aligned with the "truck shape" are the most likely to be able to engage in path capture.

todo formally express this formal claim: The edit distance between an agent's preferred path and the truck's preferred path should be minimal in order to place an agent in the class of possible dictators.

Unknowns

Unknowns turn the whole game into an epistemic game, where the possibility space blows out, and the space of available conclusions proportionally shrinks.

todo formally express this formal claim: The coalitional unknowns – the organization as the player in Wumpus World – can be derived from the participant unknowns – and are strictly a subset of the logical conjunction of all player knowledge.

todo flesh out: Information asymmetries and knowledge boundaries will evolve differently in volunteer organizations.

The trust model for running a for-profit vs a volunteer org is different.

Volunteer orgs are existentially reliant on high trust. This is because misalignment is existentially costly.
People will operationalize their work according to what they believe is the purpose of the organization.
Otoh it's trivially easy to steal labor from volunteers once you've performed path-capture and found a stable enough lie to tell. This is a systematic violation of trust that works.
- Case studies: Redmine, OpenAI.

There's a bias-variance tradeoff between building an insular vs a vulnerable community.

This seems self evident to me but I will have to unpack it.

Federated trust is a common way that volunteer orgs solve for this hard limit.

Therefore, trust in volunteer communities usually requires a fedaration idiom.

Models for volunteer trust ought to look completely different.

You need a higher emphasis on reputational trust, because the threat of letting in common patterns of bad actor is much higher but the ability to verify is much lower.
BFT protocols for decision making are necessary to mitigate against bad actors.

The compensation model for a for-profit vs a volunteer org is different.

Expect churn in a volunteer organization to be higher.

Misalignment is a frequent phenomenon and purpose-agnostic compensation functions in part as compensating for misalignment.
Everybody's playing a stag-hare game and purpose-agnostic compensation functions in part as a mitigation for the risk of playing stag. Reliable coordination is costly.
Churn is very operationally costly.
- Nonprofits often benefit from employing at least some people in order to minimize it.
- To support a volunteer organization, you want to find ways to compensate for this phenomenon.