PROOF OF HUMAN WORKA NEW BASIS FOR CONSENSUS
SECTION 01 / THE IDEA

1.The Mother
of All Problems

Begin the introduction
THE NETWORK
Independent people. A common plan.
CONSENSUS BEGINS WITH A QUESTION.SCROLL TO EXPLORE
01THE STARTING POINT

In 1982, Leslie Lamport, Robert Shostak and Marshall Pease published The Byzantine Generals Problem. They imagined a group of generals trying to agree on a common plan while some of them might be traitors. Messages could be contradicted, altered or deliberately fabricated. How could the loyal generals ever know that what they agreed upon was really the result of an honest process?

Behind the military metaphor was a much broader problem: how can independent actors reach agreement when they cannot completely trust one another, or even the information passing between them?

Computer science eventually found remarkably powerful answers to limited versions of this question. Human society never really found a solution.

Society is itself a kind of enormous consensus system. We continuously negotiate who owns what, which rules apply, how resources are distributed and what kinds of behaviour are allowed. But there is no neat division between honest and malicious participants. People, companies and governments simply have different interests and different amounts of power with which to pursue them.

That difference matters because power tends to reproduce itself. Resources buy influence; influence helps acquire more resources. Wealth buys better information, better lawyers, better technology, greater political reach and greater ability to shape the environment in which future wealth is created. It is an old feedback loop, and much of politics consists of trying to keep it within tolerable limits.

Artificial intelligence could make that loop dramatically stronger.

For most of history, intelligence was difficult to accumulate because it came attached to individual human beings. A ruler could command thousands of people, but could not literally own thousands of minds. A company could hire engineers and scientists, but their intelligence remained embodied in people who could leave, refuse, organize or disagree.

AI changes this relationship. Intelligence itself is becoming a reproducible technological resource. Programming, scientific research, persuasion, planning, surveillance, administration, economic strategy and eventually perhaps much physical work can increasingly be performed by machines.

That may create extraordinary abundance. It may also create a new concentration mechanism. More resources buy more computation and better models. Better models create economic and strategic advantages. Those advantages produce still more resources. Intelligence, perhaps the most useful resource there is, could become part of the same self-reinforcing ownership loop.

This leads to a question that is difficult to avoid: if highly capable artificial intelligence becomes concentrated, how do ordinary people retain meaningful economic and political power?

We can ask governments to distribute that power. We can ask companies to share it. We can create public institutions, cooperatives or international agreements. All of these approaches may have value, but all leave one uncomfortable question somewhere in the system: who controls the institution doing the distribution?

Nova begins with the possibility that perhaps the gate itself can be removed.

02THE IDEA

A permissionless economy

A truly permissionless system has an unusual property. Nobody decides whether you are allowed to enter it. There is no central membership database, no authority issuing citizenship in the protocol and no server whose operator can decide that your participation no longer counts. The rules are public and the state of the system can be reconstructed and checked independently.

The problem appears immediately. Digital identities are cheap. If influence were assigned per account, one participant could create a million accounts. A permissionless system therefore needs some contribution that cannot itself be multiplied almost without cost.

Bitcoin’s answer was computation. Producing blocks requires Proof of Work, so a million identities do not give someone a million times more influence unless they also have the computation to back them.

Proof of Stake takes another route and makes existing capital the scarce resource. Proof of Personhood systems try to make unique human identity scarce instead.

Proof of Personhood is attractive for an obvious reason: if every person could somehow receive exactly one unit of influence, machines could not dominate merely by creating unlimited digital identities. The difficulty is deciding what proves that a unique human exists.

There are roughly two ways of approaching the problem.

One is to build sufficiently trusted verification infrastructure. A specialized device or process can inspect a person using cameras, depth, infrared measurements or biometrics, run liveness checks and issue a credential. This can make attacks much harder, but now the security of participation partly rests on the devices or verification providers.

The other direction is to let people verify one another through synchronized ceremonies, tests or social interactions. This moves the problem somewhere else. Somebody must define what constitutes a valid ceremony or interaction, participants must coordinate, disputes must be resolved and, without any objective proof that everyone can check independently, the resulting social layer becomes part of the security model.

There is also a deeper difficulty shared by attempts to prove humanity through ordinary digital evidence. Images can be synthesized. Voices can be synthesized. Video can be synthesized. Behaviour can be synthesized. A sufficiently capable machine can, in principle, manufacture the same kind of digital object that the verifier expects from a human.

This is why Nova starts from a different question.

It does not ask how to produce a proof that artificial intelligence can never create.

It asks whether creating the proof artificially can be made more expensive than producing it physically.

That difference sounds small at first. It changes almost everything.

FIG. 01.01The network and the gate
Different ways to make influence costly.00 / 30s
FIG. 01.01How does a system decide what one more unit of influence costs?
03A DIFFERENT KIND OF WORK

Reality

Imagine a person standing in front of one or several cameras. There is a board on a table and some objects. The person is given an instruction they have never seen before. They move the board, pick something up, move it as required, change their position, touch another object and continue for some time.

From the person's point of view, very little computation has happened. They understood an instruction and moved.

From the point of view of the physical world, an enormous amount has happened.

Their body has changed position continuously. Dozens of joints have moved together. Fingers have bent around objects. Skin and fabric have deformed. Objects have moved and rotated. Surfaces have touched and separated. Shadows and reflections have changed. Fine marks on surfaces have disappeared behind objects and returned later.

With multiple cameras recording, each would capture a different view of the same event. A depth sensor, if used, would have produced another description. A spectral sensor would have seen something else again. Future sensing technologies may measure properties that are impractical to record today.

Yet all of these observations fit together.

They do not have to be reconciled afterward. Nobody has to make the shadow agree with the hand or make the reflection agree with the object. Nobody has to remember which small feature of a surface disappeared behind a finger to reappear again later.

They agree because there was only one event.

The person did not calculate any of this. Reality did.

FIG. 01.02One event
A person performs a simple physical task.00 / 28s
FIG. 01.02Reality is the source of physical complexity.

Now try to produce the same kind of witness without allowing the event to happen.

The problem is no longer simply generating an attractive video. A synthetic producer can use every shortcut available: a physics simulator, a reusable human avatar, motion models, conventional CGI, neural rendering, learned materials, cached geometry, whatever works. None of this is forbidden.

But the resulting observations still have to agree wherever the verifier happens to look. Body motion must fit object motion. Details on surfaces have to survive changes in viewpoint and occlusion. Lighting must correspond to geometry. If several cameras observe the same event, they must all be observing the same underlying event.

The more of the physical event that becomes observable, the richer the synthetic representation has to become.

This is the basic intuition behind Nova. Physical execution gets this coherence essentially for free. Synthetic execution has to manufacture enough of it computationally.

Nova does not require the synthetic proof to be impossible.

It only requires it to be expensive.

FIG. 01.03Faking the event
An attacker has to synthesize some elements of the event.00 / 30s
FIG. 01.03The question is not whether the event can be faked. It is what the fake costs.

If the cost of an accepted physical proof is CHC, and the cheapest cost of manufacturing an accepted artificial proof is CAC, the protocol is built around a simple economic relationship:

CAC≫CHC

The difference between those costs is the resource Nova tries to convert into consensus.

04THE IDEA

Why this is not another CAPTCHA

A CAPTCHA assumes that there is some task a human can solve and a machine cannot. That distinction has repeatedly moved as machines have improved.

Nova does not need such a distinction.

An AI is perfectly welcome to generate a valid Nova proof. If it can do so, the proof is valid. The important question is how many computational resources were required to manufacture it.

The verifier is public. Its models are public. Its thresholds are public. The task-generation procedure is public. There is no secret classifier that an attacker must somehow guess.

This is intentional. Nova is not supposed to win by hiding its rules.

The security comes from making a fresh accepted witness costly to construct artificially even when the attacker knows exactly what will be checked.

Nor does Nova certify someone as human once and then allow that certification to live forever. Work is continuous. New blocks require new tasks. Those tasks are derived from fresh network state and are useful only in the context for which they were produced.

A synthetic participant therefore has to continue producing new accepted work. If it wants to compete with a million people performing physical tasks, it has to produce accepted artificial witnesses at the same aggregate rate.

The cost does not disappear after one successful imitation.

It keeps accumulating.

05COMMIT. THEN CHECK.

How do you verify reality without verifying everything?

A rich physical recording could be enormous. One camera recording several minutes of activity already produces a large amount of information. A decentralized network needs a practical way for its nodes to verify that evidence.

Nova uses a simple idea to deal with this: the evidence is committed before the verifier knows which parts of it will matter.

A miner collects many completed physical tasks and commits them cryptographically. Proof of Work then generates an unpredictable audit derived from the committed data and the block being produced.

The audit can operate at more than one level. It may first choose which task recordings have to be revealed. Within those recordings it can choose particular periods, objects, viewpoints or relationships to examine.

One check might follow the details on an object. Another might evaluate a particular hand-object contact. Another might inspect a stretch of motion around an object collision. Another might look at local deformation or a change in lighting as the participant moves.

The exact checks are not sacred. Neither are cameras. They are simply the sensing and verification technology available at a particular point in time.

The important structure is that the producer has to commit the witness before knowing which parts will be inspected.

This creates an unusual asymmetry. The physical participant simply records the event. A synthetic producer has to prepare a body of evidence coherent enough to survive an audit whose location is not known in advance. Yet the validating node only has to perform a small number of checks.

Large witness. Small audit.

FIG. 01.04The random audit
Recordings form a large field of evidence.00 / 24s
FIG. 01.04The producer prepares the whole event. The network only needs to look in a few unpredictable places.

And because the verifier is deterministic, two independent nodes examining the same evidence can reach the same result. There is no central judge deciding whether a performance looks correct and human enough.

06THE IDEA

Proof of Physical Execution

Bitcoin showed that something expensive to produce but cheap to verify could secure agreement between strangers.

Nova asks whether the expensive thing has to be computation performed inside a computer.

Perhaps it can instead be physical execution.

A real physical process has a peculiar advantage: all of its consequences happen together. The world does not separately calculate geometry, movement, contact, light, deformation and perspective. They are different descriptions of one evolving state of reality.

A synthetic system may be capable of reproducing all of them. What matters is that doing so consumes resources.

This suggests that Proof of Human Work may be one application of a broader idea: Proof of Physical Execution.

The scarce resource is the cost difference between allowing a sufficiently rich physical event to happen and computationally manufacturing an accepted witness of that event.

For humans, that possibility is particularly interesting because we are already physical participants. We do not need to own specialized computing infrastructure in order to move through the world. Gravity, light, matter and our own bodies are already available to us.

Computers may be better than humans at computation. Nova does not ask humans to win that competition.

It asks whether there is an economic advantage in not having to compute the world at all.

FIG. 01.05The witness evolves
A simple action. One camera.00 / 30s
FIG. 01.05Make the physical path the cheaper path.
07THE IDEA

What could this become?

If such a primitive works, a blockchain is only the most obvious place to begin.

A permissionless source of human-derived digital scarcity raises questions about online participation, automated populations, economic distribution, governance, public goods, ownership and eventually systems in which artificial intelligence itself becomes an important economic actor.

Exactly where that leads is not something Nova needs to decide in advance.

The more immediate question is simpler.

Can physical execution become a resource that anyone can contribute, nobody can issue and machines cannot multiply without paying a larger cost?

Nova is an attempt to build a consensus system around that possibility.

Participants receive fresh physical tasks tied to the state of the network. Their performances are recorded and committed. Proof of Work determines what evidence will be inspected. Every node can independently apply the same public verifier. Accepted physical work contributes to consensus.

There is no government issuing participation rights. There is no company maintaining the authoritative list of humans. There is no single biometric service whose approval creates membership in the economy.

And there is no assumption that artificial intelligence will remain incapable of producing convincing evidence.

The assumption is much simpler:

doing it for real should be cheaper than manufacturing enough reality to pass. This is a genuine advantage that could allow us to coordinate our control and shared ownership of vastly superior AI that this AI couldn’t easily manufacture or break.

The following sections explain how Nova turns that idea into a distributed ledger, how the physical witness is generated and audited, and how the cost difference between physical execution and synthetic production can be used as a consensus resource.

CONTINUE THE EXPLORATIONFrom idea
to protocol.
SECTION 02The distributed ledger