A Six Pager on Coordination Markets

1. Azazel

Every generation a new social theory offers a fresh “solution” to “society”, galvanizing our freshmen and the existentially unmoored. Workers of the world unite, tax the rich, check your privilege. But before we Greta ourselves into the abyss, we should less enthusiastically analyze when and how our current system - capitalism - fails or underperforms: What specifically fails when you give agents the autonomy to gather information, make judgment calls, choose actions, bear consequences?

Oh! you cry, The Prisoner’s Dilemma! Aha The Moloch!!

But the games we actually play suggest a different response, one that promises less than social-welfare optimization, but also demands less than rewiring our selfish instincts and slaying the Moloch. Namely, mechanisms for committable and actionable coordination.

Consider Rousseau’s stag hunt: Two hunters can collaborate on a stag and eat well, or each hunt a hare and eat for a few hours. Both are better off coordinating on a stag – wherein neither is incentivized to defect! – but moving from hare to stag alone makes you worse off. If they are stuck in a (hare,hare) equilibrium, the challenge is not defection from cooperation rather failure to reach coordination (stag,stag).

If Moloch is the demon of defection, this is a different demon. I call it Azazel - the deity of the wilderness, pre civilization, where humans wander alone and no associations form.

The distinction between Moloch and Azazel matters because these two demons require very different remedies. If you believe society is trapped mostly by Moloch, the natural response is to build central institutions that restrain selfish behavior. Central planning, regulation, social-justice activism, superintelligence singletons. But if the problem is mainly Azazel, the remedy changes, people’s behaviour doesn’t need to be corrected or coerced, they already want to collaborate. What we lack, in the Azazel-Staghunt framework, is “merely” mechanisms to communicate and bind shared actions.

In short, we need Coordination Markets.

2. Why Cheap Talk Is Not Enough

Coordination Markets (CMs) are a category of markets built around binding primitives for coordinated action. These primitives compose, condition on external state, respond dynamically to market activity, and settle atomically.

Improving the internet through better communication channels does not suffice, since cheap talk is non-binding. In many scenarios, moving alone is risky: hunting a stag alone risks hunger and injury, rioting alone against a despot risks death, bootstrapping a liquidity pool alone risks getting eaten alive by arbitrageurs.

The missing layer, therefore, is not communication but binding intents, otherwise known as assurance contracts (Bagnoli and Lipman 1989). An assurance contract is a primitive allowing one to express and enforce conditional commitments of the form “I will do X if N others do X.” In our stack, these conditional commitments will be called intendos.

None of this is fundamentally new. PledgeBank ran conditional commitments from 2005, but the binding there was enforced by social pressure in small groups. The Point built assurance contracts in 2007, then pivoted to Groupon, more demand aggregation than assurance contracts. Kickstarter owned assurance contracts for creative projects, but remained niche. Each of these attempts either collapsed into a single vertical or was heavily biased toward petitions and activism, which is an esoteric death sentence.

Instead, we should take a market approach to coordination. A market framing has many implications; three immediate ones: First, endogenous incentives. Participants communicate their intents and commit not (merely) because they believe in the cause, but (also) because the mechanism makes commitment individually rational.

Second, a market approach is unopinionated, it should engineer for no specific agenda or world order (e/acc, radical markets). Market rails should remain open to whatever coordination problems and causes users initiate. In fact, if CMs are to unlock even a fraction of their potential, the rails should be optimized for unopinionatedness–-design for hard resilience to pressure, extortion, manipulation. This also means the mechanism should support heterogeneous preferences and risk thresholds, rather than hard-code a single safety threshold. A tenured professor might come out of the closet provided 5 colleagues do so; an assistant professor might require 50; one LP is comfortable with a liquidity bootstrap floor of $1M; another is risk-averse and demands $10M.

Third, the mechanism should be able to condition on external state. Conditions should be able to reference on-chain data, real-world events, and other facts verifiable on the web. Intendos should also be able to condition on other intendos, chaining coordination markets into multi-stage sequences.

But before any of that, CMs must allow shielding one’s intendo. Even before the move-alone risk, being the first to signal willingness is itself a primary source of risk: revealing one’s political preferences in a hostile environment, disclosing one’s financial intention in the face of arb bots, exposing a social cause when it is still small and can be killed. Allowing for confidential intents should receive axiomatic treatment.

On the other hand, collaborating in a dead dark forest is unlikely to catch momentum. Our attention is scarce, and the design should therefore borrow from and interleave with attention markets. It must foster some social p2p dynamic and/or incentive structures that outweigh the mental load of considering one’s stance and opportunities.

3. The Mechanism

A social or financial entrepreneur believes she spotted a Stag - a certain opportunity that increases the payoff of its rational (selfish) participants. She posts a Hunt, and individual users or agents discover it and contemplate. An agent deciding to join signs cryptographically an intent to join - an intendo - conditioned on a sufficient number of others joining too. The signed intendos accumulate and Pack, in an opaque process, during which more agents opt in but also may opt out at will. Once some subset of agents crosses the threshold, satisfying internally the threshold conditions of all of its members, the Hunt snaps and executes together.

Glossary.
Stag - coordination market instance; defines target outcome, execution logic, and community eligibility.
Pack - accumulation phase; intendos cluster around a Stag, aggregate state is opaque.
Hunt - resolution, the pack hunts: a qualifying subset is found and execution triggers atomically; remaining intendos persist for future resolution.

Hunts can compose (through expressive intendos) - users switch platforms and LP positions migrate to the same venue; public endorsements commit and capital deploys to the endorsed cause - all in one atomic event.

Four properties are axiomatic for the core primitive of CMs:

i. Coordinated Atomicity. The commitments of the qualifying subset - the participants whose thresholds are met together - execute simultaneously across all participants; no partial execution no gradual commit, as this would undermine the assurance.

ii. Accumulation Opacity. No-one - neither participants nor operators - can determine how close the initiative is to activation. The threshold crosses or it does not.

iii. Capital Multiplexing. Users can co-commit the same capital across different markets (eg user backing a liquidity bootstrap and a supply lock with the same 1000$); whatever commitment activates first applies. - Crucial for capital efficiency, UX scalability, and participation in overlapping markets.

iv. Composability. The output of one market is a valid input to another, and can compose with other shared state contracts. Coordination markets as lego blocks.

Both composability and capital multiplexing are unique to crypto rails. Traditional payment services operate behind isolated APIs, you can’t lego box API calls, and you can’t pre-authorize PULL based payments without statically delegating your funds to the service provider.

Any mechanism satisfying these four properties can facilitate large-scale coordination between people and agents who share similar preferences but couldn’t, so far, safely and practically express actionable coordination. It can compose populations with different resources and risk profiles, allow them to feed each other’s assurance thresholds, and trigger downstream execution that neither population could bootstrap alone.

4. Stags in the Wild

A broad category where CMs have particular utility is escaping network effect traps. Apps that “suck but everyone’s using it”, or the media platform that’s blatantly lying but everyone’s watching it. CMs allow the public’s true preference to materialize as concrete switching plans, in formal terms, it allows to aggregate demand and to alter the focality.

Example 1: Liquidity migration. Superior DeFi machinery exists but liquidity is stuck on legacy tech, 5.7B$ on BNB for instance. LPs don’t move because moving alone means providing liquidity to an empty pool. Utilizing CMs, each LP can set a threshold for migrating to kaspa - 5M$, 50M, 500M - whatever they need to feel comfortable. Their capital stays productive on BNB until a subset’s thresholds are met and resolved. The same capital can back multiple campaigns to exit BNB to different platforms, solana kaspa tempo. Individual intendos can double-sign conflicting intendos (!) and whatever triggers first applies.

Example 2: Content platform bootstrap. Netflix and Disney+ push woke agenda, and large segments of the user base - parents - resent it. Sure HBO and Amazon exist, but not huge for family content. As a concerned parent you need to know the other people ranting on the internet are willing to actually act, not just complain. Entrepreneurs face the same obscurity from the other side: If I launch a streaming service with neutral or conservative agenda, how many of those ranting parents will actually subscribe. This pure coordination failure is solvable by CMs: Users can intendo-commit subscription fees to a new platform, and provided a sufficient number join, the platform launches and charges.

Attention markets are designed to amplify organic propagation dynamics of some underlying social network. Ideally, we would wish for coordination initiatives to propagate spontaneously, via group chats, community forums, DMs. Albeit whenever we enforce Opaque Accumulation (Axiom 2) we are shielding participants’ intendos from their peers, which goes against organic social discoverability.

Can we preserve p2p propagation while maintaining privacy and deniability? Designated Verifiable Proofs (DVPs) to the rescue. DVPs enable off-the-record messaging: they allow a prover to convince a peer, or a set of peers (Multi-DVP / MDVS), that a certain statement is correct whilst maintaining full deniability in case the proof is leaked. In practice this looks like group chat messages that are internally verifiable yet practically unleakable.

Besides attention dynamics, initiators of stags should fund the continuous computation of the state of the pack. The CM stack requires a layer of operators/searchers/solvers continuously checking for subsets of intendos that can be co-satisfied. The funding for this computation can start with the stag entrepreneur -- and migrate to a fee model sustained by the pack once it reaches scale.

Entrepreneur spots a Stag → seeds a Hunt, DAC-backed → intendos accumulate (paid if it fails) → signals propagate p2p via DVPs → Pack grows opaque → compute scans for satisfiable subset → threshold hits → Hunt snaps, atomic execution.

6. The Stack

Beyond generic support for market mechanics, the stack of coordination markets requires a few new components:

i. Intendos, the layer aggregating persistent intents of users; limited implementation of persistent intents have been implemented before, eg in CowSwap, but far from the flexibility required for CMs’ composability and multiplexing.

ii. An efficient data structure and incremental algorithmic framework to solve Pack states, namely, “what is the current maximum subset with internally-satisfying thresholds?” For a large set of intendos - including thresholds defined in number of participants or amount of capital - resolving the state belongs to the class of monotone fixed-point computation, which admits incremental algorithms of time complexity O(polylog) or O(1) amortized.

iii. A computation fee mechanism appropriate for the algorithmic framework mentioned. This component includes both metering the cost (“computation gas”) per signed intendo, as well as dictating the payment mechanism, potentially distributing it across the set.

iv. A cryptographic protocol that can support opaque accumulation of the pack. The broad family here is secure multi-party computation (sMPC), but standard interactive sMPC is incompatible with continuous evaluation over an open permissionless validator set. The mostly-noninteractive alternative is threshold FHE (thFHE), which distributes a shared public key at the outset (DKG) and runs a lightweight sMPC for the decryption phase. Feasibility requires deploying mixed-mode, hiding only essential parts of the computation.

Coordination by definition leads to large market moves and cascading effects. Any gap between pack formation and execution increases the manipulation surface and multiplies pressure points. To eliminate or minimize this gap, CMs must run on infra that satisfies censorship resistance, permissionlessness, and fairness in real time. I use real-time decentralization (RTD) as a codename for this metaproperty.

7. The Consistent Individualist

In democratic ages, the bonds of human affection are extended, but relaxed (de Tocqueville).

The internet, the ultimate egalitarian project, allowed us to connect with complete anons and discover shared ideas and interests. But it did not offer ways to move together and act on common interests. It is missing an association layer which grants dynamic, ephemeral, or context-specific communities ‘write’ permissions to the shared state of the digital space.

Project Staghunt aims to build it.

March 2026

SECTION ONE - THE INTERNET

We have one democracy in the world that is, by a large gap, the very worst.

It is in constant decay and turmoil.

It is the epicenter of almost any global drama. It amplifies and complicates many world affairs.

It has near-total free speech, constant participation, and yet almost no capacity to govern itself.

This democracy has failing institutions, arguably no institutions at all.

And still we keep returning to it, interacting with it, arguing over it, although we know this toxicity is not good for our health.

I obviously care about that democracy a great deal, so I will name it for you.

It is the internet.

We treat the internet as a finished project, but it is too young to earn this title.

Democracies take at least a century to stabilize, correct, and build institutions.

We should not expect this 43 year old democracy to have figured itself out. We should not treat it as if its shape and functionality are more or less solved.

You can argue it is not a formal democracy, but you cannot deny it governs and dominates most aspects of your private and public lives.

Yes it has free speech, but free speech should lead eventually to shared understanding and action. The internet currently lacks methods to act on agreement.

Yes you can freely interact and transact with other internet members, but you are still relying on external institutions to enforce these transactions.

If the internet is the greatest egalitarian project humanity ever came up with, it is also the one needing the strongest institutions to strive.

Yet it has none.

The WWW used to feel different.

In the late 90s early 2000s, going online was an elevating experience.

It felt like a strange form of global fraternity - people talking to strangers with a depth that is mostly gone today.

Even the pornographic corners of the early internet arguably had more attachment than much of today’s discourse.

What happened? How come the internet is no longer fun?

When Alexis de Tocqueville reflected on the new society he remarked: “in democratic ages, the bonds of human affection are extended, but relaxed”, that individuals connect with more people but those connections carry less weight.

I believe this is where our diagnosis should begin; namely, by recognizing that our digital democracy obtains the broadest extent of human bonds, but also the shallowest.

That we have built barely no institutions to remedy that.

And that we have simply not been mindful enough to notice the problem.

SECTION TWO - MOLOCH AND AZAZEL

You’ve probably heard of Moloch - the deity revered by the Canaanites 3000 years ago, and more recently by Scott Alexander in his famous essay. Moloch is the demon of defection, responsible for fear and rivalry, and incentivizing people to undercut one another even when everyone ends up worse off.

In game theory, Moloch usually appears as the Prisoner’s Dilemma. Two players would be better off cooperating, yet each has an incentive to defect. So both defect, and everyone ends up worse off.

But Scott overshot it.

Not every social failure is Moloch.

Consider another game.

Two hunters can either collaborate on hunting a stag and eat for a week, or each hunt individually a hare and eat for a few hours.

If they coordinate, both are clearly better off.

Yet if one hunter fears the other will not show up, she safely hunts the hare.

And so we both end up eating rabbits.

This is the Stag Hunt: a game where cooperation is stable and self-reinforcing once reached, yet difficult to reach, because without reliable communication and binding commitments, each hunter fears the other may not show up - and the one who trusted is left hungry and exhausted.

If Moloch is the demon of defection, this is a different demon. I sometimes call it Azazel - the deity of the wilderness, before civilization, where humans wander alone and no associations form.

The distinction between Moloch and Azazel matters because these two demons require very different remedies.

If you believe society is trapped mostly by Moloch, the natural response is to build central institutions that restrain selfish behavior.

Regulation, central planning, or activist social-justice movements to correct people’s behaviour.

But if the problem is mainly Azazel, the remedy changes completely.

People do not need to be forced to cooperate.

They already want to.

What they lack are mechanisms to communicate commitment and bind themselves to shared actions.

This difference shapes our political imagination - especially yours, as Oxford students.

Workers of the world, unite.

Occupy Wall Street!

Eat the rich.

Every few decades a new slogan galvanizes your hearts.

The desire is real, you identify that the state of affairs can be improved if people worked together. But the game is not rigged, and human agency is not corrupt. You can still make a dramatic impact by working with that, and building institutions that better the equilibrium selection of the free markets game.

Do not waste your hearts on slogans that try to replace the self-interested individual with some collective virtue. This always leads down the road to serfdom.

In other words, solve the Stag Hunt.

SECTION THREE

The real problem is assurance - “I move only if others move.”

Hong Kong protests, Libertarian Party support, liquidity migration.

What mechanism would fix this?

A. Assurance, atomic action.

B. Opacity, neutrality.

SECTION FOUR

We derived that the internet democracy desperately needs rails for assurance and coordination.

Requirement recap:

Autonomously, no dictator entity, shared language, enforceable rules, programmable actions, shielding opaque intents.

Everything that was built in crypto in the last two decades, from stateless money base, to programmable money, to encryption techniques, are the building blocks the digital democracy needs.

Crypto hasn’t necessarily recognized this yet. The industry is still in a bear cycle mood, mourning its old narratives and the lost market caps. But the real mission is still ahead of us.

Crypto should be building coordination markets - what I like to call Project Stag Hunt.

SECTION FIVE - HOW A COORDINATION MARKET RUNS

Let me show how a coordination market actually runs. I will use three simple terms - Stag, Pack, and Hunt.

5.1 STAG

A stag is spotted - a better equilibrium.

Example proposal - “Move to a new social platform if enough users commit, and pay $5/month.”

No one moves yet. Everyone waits to see whether others will move.

5.2 PACK

Users join the pack by submitting conditional commitments. “I migrate if at least N others migrate.”

Each participant sets their own threshold. Commitments accumulate privately. No one risks moving alone.

5.3 HUNT

Eventually a subset of commitments satisfies all their thresholds.

At that moment the pack hunts.

The migration executes atomically.

Accounts activate, communities appear, and the network launches with users already there.

5.4 COMPOSABILITY

These hunts are composable - like Lego blocks.

A set of investors might sign commitments: “I invest $10 million if this hunt completes.”

When the hunt resolves, all of these commitments execute atomically.

Users move, capital deploys, infrastructure appears.

5.5 IMPORTANCE OF EMERGENT BEHAVIOUR

Nobody specific coordinated this or engineered society for one specific opportunity.

Decentralized.

No Leviathan or superintelligent singleton.

SECTION SIX - THE BUILDERS

In 1980, the political scientist Langdon Winner posed a question: Do artifacts have politics?

He described a system of bridges on New York’s Long Island - built too low for public buses to reach the beach. They were planned by Robert Moses, a high-status New York urban planner who designed a road that only car owners could use.

Whether this was conscious discrimination or an oversight matters little.

Artifacts are built in the image of their creators.

The cypherpunk pioneers are the Robert Moses of the internet. They built the essential infrastructure for the digital egalitarian project. TCP/IP, encryption, stateless money, programmable contracts, self-custody.

But when it comes to civic tech - community coordination and collective action - the contributions are mostly fringe and disconnected.

Webs of trust, liquid democracy, DAOs - these artifacts were built with the right intention but by the wrong type of builders: brilliant introverts, paranoid, and more comfortable minimizing trust than organizing cooperation.

Hayek wrote: “The consistent individualist ought to be an enthusiastic supporter of voluntary associations”. Many cypherpunks might agree with this in principle, but rarely in temperament.

The internet rails - and definitely crypto - could not have been built by any other mindset or culture. Adversarial trustless personalities are the ultimate builders of the bare backbone of the internet.

But now the roads need to lead somewhere, and a different class of builders should take the reins.

I hope some of you here at Oxford will lead it.