Coordination is the lifeblood of democracy. People finding common cause, building coalitions, amplifying shared interests—this is how collective action happens. Without it, we're just isolated individuals shouting into the void.
Coordination is also how democracy gets captured. People finding common cause against the public interest, building coalitions to extract value, amplifying shared interests at everyone else's expense—this is how oligarchy happens. With enough of it, collective decisions become private deals wearing democratic costumes.
The line between legitimate coalition and illegitimate cartel isn't always obvious. Sometimes it isn't a line at all.
The Cooperation Problem
Game theory offers a useful frame here.
In a healthy democratic system, participants face a cooperation problem: how do we coordinate enough to make collective decisions while preventing any subset from dominating? The goal is cooperation toward shared outcomes, not cooperation to capture outcomes.
Cartels flip this. A cartel is a group that solves their internal cooperation problem in order to exploit everyone outside the group. They coordinate beautifully with each other. That coordination extracts value from non-members.
OPEC is a cartel. Member nations cooperate on production quotas so they can collectively charge higher prices than competition would allow. Their internal coordination is impressive. Its purpose is external exploitation.
Voting cartels work the same way. Members agree to vote as a bloc, concentrating influence that would otherwise be dispersed. The bloc gets outsized power. Everyone else gets diluted.
How Cartels Form
Cartels emerge when coordination costs drop below coordination benefits.
In traditional elections, forming a voting cartel is hard. You need to identify potential members, communicate secretly, establish trust, coordinate timing, and enforce compliance—all without detection. The logistics limit scale.
Online systems collapse these barriers. Communication is instant and cheap. Pseudonymity can provide cover. Smart contracts can enforce agreements automatically. Discord servers can coordinate thousands of participants in real time.
The same tools that enable legitimate organizing also enable capture. A neighbourhood association coordinating votes on park funding uses the same infrastructure as a bad actor coordinating votes to extract treasury funds. From the outside, the patterns look identical.
Delegation and Its Discontents
Many modern voting systems include delegation: the ability to transfer your voting power to someone else. The theory is that busy participants can still have voice by choosing representatives who share their values. Delegation scales participation.
Delegation also scales capture.
If I convince a thousand people to delegate to me, I now control a thousand votes. I might use this power responsibly, reflecting the interests of those who trusted me. I might also use it for personal benefit, confident that most delegators aren't paying close attention.
This is the principal-agent problem in its purest form. Delegators want representation. Delegates want... what? Maybe representation. Maybe influence. Maybe deals with other delegates. The interests may align or diverge, and delegators rarely have visibility into which.
Worse, delegation concentrates targeting. If you want to corrupt a thousand votes through direct bribery, you need a thousand transactions. If those votes are delegated, you need one transaction with one delegate. Delegation creates high-value targets for corruption.
Liquid Democracy's Broken Promise
Liquid democracy was supposed to solve this.
The idea: delegation should be fluid, revocable, and transitive. You delegate to Alice. If Alice's votes stop reflecting your values, you withdraw and delegate to Bob instead—or vote directly. The threat of exit disciplines delegates. Power stays accountable.
In practice, liquid democracy created delegation marketplaces with their own dysfunctions.
Attention asymmetry - Delegators chose liquid democracy precisely because they don't have time to pay attention. How will they notice when their delegate drifts? The feature that makes delegation attractive—reduced attention burden—undermines the accountability mechanism.
Transitivity exploitation - You delegate to Alice. Alice delegates to Charlie. You've never heard of Charlie, but Charlie now controls your vote. Transitive delegation chains can concentrate power in ways no individual delegator intended or understands.
Delegate oligopolies - Network effects favor prominent delegates. New delegators choose established names because they're visible and seem trustworthy. Established delegates accumulate more power, becoming more visible, attracting more delegators. The system converges toward a small set of power brokers regardless of their ongoing quality.
Vote selling by proxy - Direct vote selling is detectable: money flows to voters. But what if delegates sell their accumulated voting power? The transaction looks like a "donation" or "consulting fee" to a community leader. Individual delegators never see cash; they just notice their delegate voting differently.
Liquid democracy made these dynamics more sophisticated, not less present.
Cartel Detection Is Hard
How do you identify cartel behavior?
Some patterns are suspicious: groups that always vote identically, accounts created around the same time that coordinate on every proposal, sudden vote swings that suggest pre-planned deployment.
But all of these have innocent explanations too. Political parties vote together—is that a cartel? Communities with shared values coordinate—should they be penalized? Sudden swings might reflect genuine persuasion or new information.
The difference between "coalition" and "cartel" often comes down to intent and effect, neither of which are directly observable. A group voting together because they genuinely share interests is democracy working. A group voting together because they've agreed to split the proceeds is democracy failing. The votes look the same.
This is why pure detection is insufficient. You also need mechanisms that make cartel formation less rewarding, even if you can't identify specific cartels.
Mechanism Design Against Collusion
Several approaches try to make coordination attacks less profitable:
Quadratic funding and voting - We discussed quadratic mechanisms as a defense against plutocracy. They also resist cartels. If coordinated voting has diminishing returns, the incentive to form blocs decreases. A cartel of 100 members doesn't get 100x influence—it might get 10x. This doesn't eliminate coordination advantages, but it compresses them.
Pairwise coordination penalties - Some mechanism designs penalize votes that correlate too closely. If your votes always match another account's votes, both votes get downweighted. This creates incentives to differentiate even within coalitions. Of course, it also penalizes genuine agreement, and sophisticated cartels can add noise to evade detection.
Randomized participant subsets - Instead of letting everyone vote on everything, randomly assign proposals to voter subsets. Cartels can't concentrate their full power on any single decision because they don't know which members will be eligible. This trades comprehensiveness for manipulation resistance.
Conviction mechanisms - Requiring sustained commitment over time raises cartel coordination costs. It's easy to get 100 people to click the same button once. It's harder to get them to maintain locked votes through weeks of changing circumstances. Patience costs more for groups than for individuals.
Identity and reputation costs - If participating in governance requires reputation that takes time to build and can be lost, cartel members risk something by coordinating. Getting caught in cartel behavior could mean losing accumulated standing. Whether this deters coordination depends on enforcement credibility.
The Dark Forest
Here's the uncomfortable truth: the best cartels never get caught.
A well-designed cartel doesn't vote identically—it coordinates on which votes to concentrate force on while maintaining apparent independence otherwise. It doesn't use traceable communication channels. It doesn't split proceeds in detectable ways.
The cartels we identify are the incompetent ones. The sophisticated versions look like organic community consensus. We can't know how much observed agreement reflects genuine shared values versus hidden coordination. This uncertainty is itself corrosive to democratic trust.
This is the "dark forest" theory applied to governance: the absence of visible predators doesn't mean safety. It might mean predators too competent to be visible.
Transparency As Partial Defense
If coordination is happening regardless, what's the argument for making it visible?
Open coordination—explicit coalitions, public voting blocs, acknowledged interest groups—at least allows counter-coordination. If you know the landlord association votes together on housing policy, tenant advocates can organize their own bloc. The playing field isn't level, but the game is legible.
Hidden coordination advantages those with resources to coordinate secretly while leaving less-resourced groups disorganized. Forcing coordination into the open might paradoxically help those with less power, because organizing publicly is cheaper than organizing covertly.
This is a weak defense. It doesn't eliminate cartel advantages. But democratic legitimacy might survive open factions competing in ways it can't survive hidden capture.
The Federation Question
Scale matters enormously here.
In a small community—a few hundred people who know each other—cartel formation is constrained by social costs. If you're caught coordinating against your neighbours' interests, you face consequences beyond the voting system. Reputation, relationships, and future cooperation are all at stake.
In large anonymous systems—millions of participants who'll never meet—these social constraints vanish. Cartel formation becomes a pure coordination problem with no interpersonal costs.
This suggests that voting systems should perhaps be federated rather than unified. Keep decision-making at scales where social constraints still operate. Let communities govern themselves rather than aggregating into anonymous masses where cartels thrive.
But federation creates its own problems: capture of small communities by local elites, inconsistency across jurisdictions, barriers to participation for mobile populations. There's no escape from tradeoffs.
What Defenders Should Remember
Collusion isn't an aberration—it's the default behavior of rational actors in systems where coordination pays. Expecting participants not to coordinate is expecting them to leave value on the table.
Design accordingly. Assume cartels will form. Build systems where cartel advantages are minimized, where coordination benefits compress rather than compound, where detection has some chance even against sophisticated actors.
And maintain humility. The cartel you're worried about might be less dangerous than the cartel you haven't imagined. The defenses you implement might be gamed in ways you can't foresee. Security is an ongoing process, not a destination.