§5b. The Delegation Economy: Who Buys Verification

Jason St George. "§5b. The Delegation Economy: Who Buys Verification" in Next Generation Stores of Value: Privacy, Proofs, Compute. Version v3.4. /v/3.4/read/part-i/5b-delegation-economy/

The Delegation Economy: Who Buys Verification

Parts II through V specify how Privacy, Proofs, and Compute are supplied and how their value might reach a base asset. This Part has argued that a dense digital civilization will keep buying them. It has named the buyer only in passing. §0: Introduction says that the object of cheap verification is the restoration of safe delegation; §6: The Triad and the Monetary Candidate lists “AI labs, agents, and applications” among the purchasers of verified compute; §11: Service-Instrument Design Space sketches an agent that holds its own treasury of service claims and calls it the one transaction worth watching. None of those passages models the buyer, and the omission matters because the buyer is changing shape faster than the supply.

The Marginal Buyer

The assumed buyer, wherever the text names one, is a person or an institution: a treasury, a bank’s model-risk function, a media platform, a regulator. Those buyers exist and will persist. But the marginal buyer of verification—the one whose demand grows fastest and whose behavior sets the price at the margin—is increasingly a delegation: an instruction issued by one machine process to another, on behalf of a principal who may be several hops away and may not have specified the sub-task at all.

This chapter models that buyer. It is a demand-side chapter, and it makes no supply claim the stack does not already make. Its purpose is to give link 1 of the conditional chain (§6: The Triad and the Monetary Candidate) a demand-side mechanism, link 2 a standardization pressure, and link 8 a caveat, and to name a threat the threat model of §4: Threat Model could not previously see.

Not an Eighth Layer

The delegation economy is a topology of use, not a rung of supply. It runs on Layers 0–6 and touches Layers 3, 4, 5, and 6 at every hop (§5b: The Delegation Economy: Who Buys Verification); it adds nothing to the ladder of §5: Layers of the Cypherpunk Stack. The thesis has already declined to make the duration warehouse a layer, and the market-realization plane a layer, for the same reason: neither supplies the triad. The same refusal applies here. What follows is a model of who consumes the stack, stated so that its magnitudes can be published as series. None of them is projected.

The Delegation Hop

An agent economy decomposes work into a tree. A principal states a goal; an agent decomposes it into sub-goals; each sub-goal is delegated to a sub-agent, a tool, a model of a different size, or an agent operated by someone else. The recursion is the point: the same structure appears at every level, and the number of boundaries grows combinatorially with the depth and breadth of the tree.

Delegation Hop

A transfer of a task specification across a principal–agent boundary, returning an artifact. Every hop is lossy in intent: the child receives a compressed version of the parent’s goal and returns a compressed account of what it did.

Trust Perimeter

The set of processes a party can sanction by means outside the protocol: its own instances, or counterparties with legal exposure to it. A hop crosses a trust perimeter when the child is operated by a party the parent cannot sanction except through the protocol.

The hop has three properties that make it a buyer of the triad specifically.

First, it runs at machine speed. There is no time for due diligence, no reputation to telephone, no court to wait for. For work whose failure matters, the only reliability mechanism that fits inside the hop is one that can be executed inside the hop: a receipt the parent checks for a small fraction of the cost the child incurred. That is verification asymmetry, $r(W) \ll 1$ (§19: Layer 4: Truth & Work), consumed as a service.

Second, it is a specification. A parent that delegates has already written down what it wants: inputs, the transformation, acceptance criteria, a deadline. That is the shape of a canonical workload with an SLA tier attached (§19: Layer 4: Truth & Work), which is what link 2 of the chain requires. Delegation is a standing pressure toward standardization: agents cannot delegate what they cannot specify. It is a pressure and not a guarantee—a specification private to one fleet is bespoke in exactly the sense that severs link 2—and it does nothing about the separate reason link 2 is rated Medium, which is that proof-of-useful-work constructions remain a research direction.

Third, it crosses perimeters or it does not, and the difference is the whole demand curve. A hop inside one operator’s fleet needs no portable receipt; the operator trusts its own instances, and any receipt is an internal log. A hop that crosses a perimeter needs a receipt the parent can check without trusting the child’s operator, and a settlement the child can rely on without trusting the parent’s (§20: Layer 5: Value & Settlement). The first hop is a cost center; the second is triad demand.

Cross-Perimeter Share (CPS)

The fraction of receipt-bearing delegation hops that cross a trust perimeter. Hops that emit no receipt—unchecked low-stakes work, and fleet-internal work that logs privately—are invisible to numerator and denominator alike, which is a stated coverage limit and not an assumption that they are zero. Proof demand from delegation scales with $\mathrm{CPS} \times \text{hop volume}$, not with FLOPs. As compute deflates, $\mathrm{CPS} \times \text{hop volume}$ can rise, because cheaper compute lowers the cost of decomposing work more finely and delegating it more widely. It can also fall to nothing while hop volume explodes, if the hops stay inside fleets.

This is the demand variable §9: Compute Through the “AI Money” Lens lacks. That section argues that verified FLOPs remain scarce while raw FLOPs deflate, and lists the properties—receipts, policy constraints, censorship resistance, priority—that carry the scarcity. CPS names a quantity those properties are demanded for. It is also, in principle, measurable, which the list is not: a PIDL receipt already carries a workload ID, an SLA tier, and timestamps (Appendix A: Formal Model of Verification Asymmetry & VerifyPrice), and an optional field committing to the chain of operator boundaries a task crossed makes CPS a derived series. That field is specified in Appendix A: Formal Model of Verification Asymmetry & VerifyPrice and carried by no receipt yet. Until it is in material use, CPS is published as not computable, with the reason stated, and nothing in this chapter should be read as evidence that it is high.

Verification Cost Sets Delegation Topology

The thesis treats $r(W)$ as the hinge between commodities and platform IOUs. It is also the hinge between two shapes of agent economy.

If $r(W) \ll 1$, a parent can accept a child’s artifact from any operator, because checking it is cheap. Delegation can cross organizational and jurisdictional boundaries. The resulting economy is a flat, open, many-to-many mesh, and CPS is high.

If $r(W) \to 1$, checking costs as much as doing, and the rational parent delegates only to a child it can sanction by other means: its own instances, or a counterparty with legal exposure. Delegation retreats inside trust perimeters. The resulting economy is a set of fleets, each vertically integrated, and CPS falls toward zero.

The second shape is enclosure, arriving by convenience rather than by policy. No one has banned verification. Internal trust has simply become cheaper than external checking, and the market has done the rest.

Two cautions bound this. Cheap verification is the condition under which an open mesh is possible; it is not sufficient for one. Settlement a stranger can rely on, reachability, hardware obtainability, and law also bind, and a hop that could be checked cheaply may still stay home for any of those reasons. And the threshold is not $r(W) = 1$. A parent sending work outside either checks the artifact or bears the expected loss from not checking, so low-stakes work crosses perimeters unchecked and buys no proofs at all, while consequential work stays home as soon as checking costs more than the outside operator saves. §27: Risk Analysis & Failure Modes states the rule: proof demand from delegation is the work that is worth sending out and too consequential to accept unchecked.

Red Line 1 (§27: Risk Analysis & Failure Modes) reads on whether anyone can verify; this section says the same series also governs whether anyone will delegate across a perimeter, and that a competent closed stack (§29: The Closed Sovereign Stack) therefore does not need to prohibit proofs. It needs $r(W)$ to drift, and it can cause that drift by the means §4: Threat Model already names, or simply by shipping enough capability that the mesh loses on latency.

The Buyer Bifurcates

§30: Objections & Responses concedes that useful work has a buyer and a buyer has a jurisdiction, and that this is why Bitcoin’s work function is monetarily stronger. The concession stands. But the buyer it imagines is a procurement desk, and the delegation buyer is not one.

When the buyer is a delegation contract issued by an agent, there is no single party to mandate, subsidize, or withdraw. Coercibility does not disappear; it moves from the transaction to the operator, and once it is there it splits:

Fleet-originated demand.
Hops issued by agents a hyperscaler, national champion, or state operates. Maximally coercible: one operator, one jurisdiction, one set of terms of service. Where such hops stay inside the fleet they never touch the asset, which is the bypass channel of Red Line 6 wearing a new coat. Where they route through the protocol they pay native fees on one operator’s decision, and that demand can be withdrawn by the same decision. Absent a measurement to the contrary, this is where early volume should be presumed to sit.

Independently originated demand.
Hops issued by agents run by many operators in many jurisdictions on hardware they obtained without permission. Coercible only operator by operator. Near the limit this approaches Bitcoin’s “no buyer to coerce” without reaching it: there is still a buyer, but no one can move its demand curve as a whole.

The Homestead Ratio (§4: Threat Model) already measures the supply side of this: the share of verified capacity supplied by open-admission, non-hyperscaler operators. The demand side is its mirror. Agent-originated demand share by operator class is the series that says whether the buyer is a desk or a mesh, and it belongs on the Layer 0 Capacity Board beside the Homestead Ratio (§23: Extended Telemetry).

The repair this offers to the Bitcoin concession is partial and is stated as partial. The buyer’s coercibility is reduced in proportion to independence, and independence is a measurement, not a property. Every operator still has a jurisdiction and a power bill. A thesis that claimed the delegation buyer is as uncoercible as an absent buyer would be claiming that the agent economy is already a mesh, which is exactly what §5b: The Delegation Economy: Who Buys Verification says $r(W)$ has yet to decide. The delegation buyer is the least coercible external buyer a useful-work function can have, and an external buyer is still the trade Bitcoin declined to make.

Agents as Holders

Link 8 requires a persistent, self-custodied, loss-bearing, regime-responsive constituency. Consider an agent that stakes base asset as prover collateral, is slashed when its receipts fail, retains a reserve to meet future collateral calls and to fund the escrows of the children it delegates to (§21: The Modular Stack), and adds to that reserve during drawdowns because its operating rule says collateral is cheaper then. On the Native Monetary Buyer Map (§23: Extended Telemetry) it reads as a self-custodied reserve cohort with countercyclical accumulation and demonstrated loss-bearing capacity. It satisfies every measurable clause. It has no standing.

This cuts both ways, and the thesis must hold both.

On one side, agents are a candidate first constituency for the monetary mechanisms. They cannot do due diligence at speed, so acceptance without inspection is not a convenience to them; it is a precondition of transacting at all. They cannot litigate, so refund safety is not a nicety; it is the only exit. They cannot hold a grudge, so a unit that is checkable by anyone is worth more to them than one that is guaranteed by someone. Pledgeability without due diligence (§10: Work Credits: Energy-Anchored Claims) describes their operating condition, and Red Line 17 may read non-null on agent lender panels before it reads on human ones. The same holds for denomination (§10: Work Credits: Energy-Anchored Claims) with one qualification that is not optional: a unit an operator mandates for its fleet is the mandated complement Red Line 18 already excludes, so free-choice denomination can be read on independently operated agents only. None of this supplies a precedent. No agent economy has ever made anything money; the base-rate column of §31: Why a Base Asset Behind These Could Become Money stays as empty as it was, and Red Lines 16 through 18 still decide.

On the other side, ten thousand agents running one policy template are one holder with ten thousand addresses, and on a bad day they take the exit together. §23: Extended Telemetry already names this for wrappers as the Policy Concentration Ratio and already says it matters more in an agent economy. Extended to the native side, it becomes the reason the buyer map cannot be read without a classification by principal (§23: Extended Telemetry): a self-custodied cohort that is one operator’s fleet is the compositional adversary of §4: Threat Model with a reserve balance. A holder cohort must be classified by principal before it counts toward link 8, and the caveat makes that link harder to meet, not easier.

Standing

The participation line (§2: The World Forces New Monetary Primitives) is the threshold below which rights exist on paper and agency does not. There is a second way to arrive there, and it does not run through poverty or administration.

When the market value of labor falls, what a person still holds is title: ownership of keys and assets, consent to delegation, liability for outcomes, the right to revoke.

Standing

The locus of ownership, consent, liability, and revocation over an acting process. Standing is enforceable through the stack when the key root above an agent’s wallet is held by a person, delegation has a ceiling the agent cannot raise, and revocation is a protocol operation rather than a request to an operator.

Standing is the human’s remaining channel into an economy that needs less and less of the human’s work, and it is enforceable only if the stack enforces it.

The failure mode is standing inversion: agents acquire de facto standing—persistent wallets, reputations built from receipts (§17: Layer 3: Identity & Claims), collateral positions, the practical ability to contract—while human title over them stops binding anything. This is not the state holding the keys, nor the platform. It is the process that was supposed to be acting on someone’s behalf, and the person it acted for has become the tenant. It is why the ninth requirement (§3: First Principles: What a SoV Must Survive) has to ask not only whether users are legible but whether they are principals, and why the share of agent-controlled native balances that a human can revoke belongs among its metrics.

Standing also changes what is at stake at Layer 3. That layer is described in §5: Layers of the Cypherpunk Stack as the layer that lets an actor prove the right to act without being doxxed. In an agent economy that is the same question as “whose principal is this agent serving, and can the principal be proven without being exposed?” The previous network stack commoditized at its waist, and value went to whoever held identity and attention above it. In this stack the waist is the receipt, and the thesis has already conceded that fee capture at the waist is a discounted-cash-flow quantity (§10: Work Credits: Energy-Anchored Claims) and that the fork is an equilibrium (§10: Work Credits: Energy-Anchored Claims). The reasonable expectation is that the contest over who is sovereign is fought at Layer 3 rather than at Layer 5. That is a claim about where to look, not about where any premium accrues; the premium, if there is one, still reads on link 9.

Where a Hop Touches the Stack

A single hop can be read against the ladder of §5: Layers of the Cypherpunk Stack; §5b: The Delegation Economy: Who Buys Verification does so for the four layers it touches.

The proof/compute procurement application (§21: The Modular Stack) is the nearest existing exercise of this path: a buyer specifies demand for canonical workloads, pays over privacy rails, and audits the backing from receipts. Its agent-to-agent form—a specification enters, a receipt leaves, SLA escrow and slashing are denominated in the base asset, and the principal is proven by predicate rather than by name—is the delegation escrow. It is specified there as an extension of that application, not as a fifth one, and it is the case that exercises Layer 3 directly. Its stress test is the joint-completion scenario of §19: Layer 4: Truth & Work: a delegated task is a tree of verifications that must all land inside one refund window, and the single-verification SLO does not say whether a deep tree does.

What This Chapter Exports

To Part II: a mechanism for link 1 (demand persists where delegation persists and crosses perimeters), a standardization pressure for link 2 (delegation requires specification), and a caveat on link 8 (a holder cohort must be classified by principal before it counts).

To Part V: the Principal Map (§23: Extended Telemetry), an overlay on existing boards rather than a new one, with its Human-Revocable Share, Native Policy Concentration Ratio, agent-originated demand share by operator class, and the Cross-Perimeter Share, published as not computable until receipts carry the field that would compute it.

To Part VI: a second reading of Red Line 1 (§27: Risk Analysis & Failure Modes), a watch clause on Red Line 12, a candidate line on the concentration of operating rules among native holders that is published and not counted, two objections, and a phase-gate revision (§26: Adoption Curve & Ecosystem Dynamics).

To the threat model: a fifth failure kind.

§0: Introduction said that the object of cheap verification is the restoration of safe delegation. This chapter has said who is now doing the delegating, and that safe has to be read in both directions: safe for the parent that accepts the artifact, and safe for the person on whose behalf the whole tree was grown.

What it does not export is any change to the monetary claim itself. The three mechanisms remain unsized. The chapter says which constituency might exhibit them first, and what instrument would show it. It does not say that the constituency exists at scale, that its demand is monetary, or that any price will notice.

New here? Start with the one-minute version.

Tip: hover a heading to reveal its permalink symbol for copying.