Research commentary · Digital Scarcity

A seat is not a credit.

A firm buys ten permissions. Does that mean ten people may work at once, or that ten pieces of work may ever be completed? The same number can describe two different economic goods. A system that cannot tell them apart has not specified what it makes scarce.

One count, two promises

Consider a hypothetical design service. Under one contract, a customer buys ten concurrent seats for a month. A worker occupies a seat while using the service. Once that use has ended and the seat has been released, another worker can occupy it. Under a second contract, the customer buys ten rendering credits. Each completed rendering consumes a credit. Finishing the job does not put the credit back.

Both products can have a counter initially showing ten. Their histories are different. If ten workers finish ten jobs, the seat contract can again have ten available places. The credit contract has no unspent credits. Returning its counter to ten would issue another ten claims, not merely release occupied capacity.

This is a practical distinction for digital scarcity. Scarcity need not mean that an object can be used only once. It can mean that a specified capacity cannot be used by more than its permitted holders at the same time. Conversely, something described as access may really be a finite stock of consumable claims. The object must be identified before its transfer, price or enforcement can be assessed.

The familiar mechanism, and its limit

The acquire-and-release pattern is established computer science, not a new invention. Python's official semaphore documentation describes a counter reduced on acquisition and increased on release, with acquisition waiting when no capacity remains. A bounded semaphore checks attempts to raise the counter above its initial limit. The documentation also cautions against relying on a particular order in which waiting threads resume.

That is a useful local model of a concurrent seat. It is not by itself a digital-property system. A process-local counter does not determine who owns the entitlement, coordinate independent suppliers, prove that a remote user stopped working or settle an ownership dispute. Those additional promises need their own mechanisms. Calling the counter a token does not supply them.

The event that restores supply

A reusable permit has a release event. A consumable claim has a consumption event. A refund or replacement is a third event, authorised under its own conditions. Keeping those events separate prevents an apparently harmless maintenance operation from creating extra supply.

For the ten-seat service, a simple invariant is that the number of currently occupied seats plus the number available equals ten. A successful acquisition moves one seat from available to occupied. A valid release moves that same seat back. An attempted second release of the same occupancy must not create an eleventh place.

For ten credits, the corresponding accounting is different: unspent credits plus completed consumptions equals ten, unless the issuer has explicitly recorded a refund or new issue. A completion moves a unit into the consumed category. It does not become availability. If a job fails before the contractual consumption point, the credit may remain unspent; if it fails afterwards, compensation depends on the stated contract.

Even an occupancy identifier is not enough to prove remote cessation. Suppose a device becomes unreachable while using a seat. The service cannot infer from silence alone that the work has ended. Reassigning the seat while the old device can continue would violate the concurrency promise. A timeout can be part of an enforceable expiry design, but only where continued use is actually stopped or bounded accordingly. Otherwise the operator has changed a safety promise into a hope.

A small, explicit accounting example

Take two reusable seats and two consumable credits. Alice acquires a seat; Bob acquires the other. A third acquisition is refused. Alice's verified release allows Carol to acquire that place. Three people have now been served, but no more than two have occupied seats at once.

With credits, Alice completes one job and Bob completes another. Carol cannot complete a third credit-funded job without an additional issue or a refund that is justified under the contract. The system may have abundant idle computing capacity and still owe Carol no service under the exhausted two-credit purchase. Physical availability and purchased entitlement are separate questions.

The executable illustration accompanying this article enumerates short acquisition-and-release sequences and short credit-consumption sequences. It checks capacity bounds and unique consumption identifiers, and verifies that duplicate releases cannot replenish capacity. These are tests of a small sequential model, not a security certification of a distributed product. The useful result is the visible difference between reusing a unit and issuing another unit.

Why the distinction changes the price

The buyer of a concurrent seat pays for a ceiling on simultaneous use over a stated period. Its value depends on how often demand overlaps, how long each use lasts and whether the service is available when required. A credit buyer pays for a number of qualifying consumptions. Its value depends on what counts as a job, whether jobs differ in resource demand and when the claim expires.

As a purely illustrative calculation, two seats used in five non-overlapping rounds can accommodate ten jobs. Two credits can pay for only two jobs. But five rounds is an assumption about scheduling and duration, not an inherent ten-job guarantee. If every job takes the entire available period, two seats serve only two jobs. If all ten requests arrive together and cannot wait, the same seat purchase is insufficient even when its average utilisation would have been low.

Consequently, a comparison of prices per “token” can be meaningless. A fair comparison specifies the purchased unit, the time window, the permitted overlap, the service obligation and the remedy for non-performance. It also distinguishes the entitlement from the resources needed to fulfil it. Selling ten seats does not prove that a supplier has enough processors, bandwidth or staff to support ten satisfactory simultaneous uses.

A stable unit is an institutional commitment

My published research on governance, commitment and rule change in public blockchains distinguishes open admission from the authority to alter the rules on which people rely. It examines the exposure of chain-specific investment to later changes in those rules. The seat-and-credit example applies that commitment question to a particular service contract; it is not a mechanism reported or empirically tested in that paper.

A purchaser may build a workflow around reusable seats. If the supplier later treats every completed use as permanent consumption, the same displayed quantity no longer represents the same bargain. The reverse change can expose a supplier or other holders to more usage than the original promises covered. A ledger may accurately record the renamed units while the economically important rule has changed.

A credible offering therefore states not just how many units exist, but what keeps a unit occupied, what consumes it, what restores it and who may change those definitions. Ownership transfer must preserve the chosen semantics. Moving a credit to another holder must not turn it into a reusable seat; moving a seat must not leave an old occupancy able to continue beyond the agreed handover.

Build the promise before counting it

The opportunity is not to make every digital service artificially scarce in the same way. It is to let different services make precise, testable commitments. Reusable licences, metered computation, reservations and consumable permissions can support different businesses because they restrict different things.

Before comparing implementation speed or transaction fees, ask a simpler question: what event makes the unit available again? If the answer is release, test release. If the answer is that nothing does, test consumption and separate every subsequent issue. The difference is small enough to fit in a counter and important enough to change the product being sold.

Inspect the finite accounting model.