Deferred-primitives pressure

audience: contributors

The blueprint defers enforceable coalition-level primitives: no protocol-level fees, no in-protocol membership requirements, no admission filters the substrate itself imposes, no adjudication machinery. Topology-intro names this the deferred-primitives space. The CLAUDE.md conventions flag any drift toward it as a bug.

This page names the site on the blueprint’s surface where that drift is most likely to arrive: the web2 compute-bridge walkthrough. Compute and Randomness both have a market-for-supply structure, which is what makes the blueprint useful for the work the book is about, and also what makes them the obvious surfaces for regulatory pressure to attach to.

Why compute sits on the edge

Compute is the basic service that touches the outside world at the closest range. Atlas, Almanac, and Chronicle are internal to the coalition — a directory, a clock beacon, an audit log. Compute is not: it bridges to cloud providers and bare-metal hosts, and those two domains already carry regulation.

Export control on AI accelerators. AI-specific workload reporting — US EO 14110, EU AI Act Article 51 and onward, the UK regulator frameworks. GPU licensure. Jurisdictional cloud-residency rules. Whatever of these applies to a compute-bridge operator at their jurisdiction of incorporation applies whether or not the substrate knows about it.

Compute also carries settlement evidence. The requester hands the module an opaque hash; off- module that hash usually resolves to an on-chain payment, which brings money-transmitter frameworks, FATF travel-rule regimes, and per-session reporting thresholds into play once values cross a declared floor.

Compute hosts third-party workloads. Cloud providers have been subpoenaed for workload records; a compute-bridge operator served with a subpoena for its ComputeLog stream cannot refuse on the grounds that shuffle anonymisation hid the requester — the log still names the provider, the window, and the usage.

Reputation organisms that commit binding scores about named providers sit adjacent to defamation law and, depending on market share, competition law. A coalition that pins a reputation floor in its member validators is running a light-touch licensure regime in all but name.

Randomness inherits the same exposure through the same market-for-supply structure. Beacon providers often operate in already-regulated domains (lottery, gambling-adjacent randomness, public-sector randomness). The quirks are different from Compute’s, the shape of the creep is the same.

What the creep looks like

Each vector below usually arrives as a small, technical improvement. Stacked, they rebuild a regulator on top of the substrate.

• Requester due diligence. “A small allowlist on ComputeRequest to prevent abuse.” The allowlist becomes protocol-level admission; providers end up compelled to enforce it whether or not their jurisdiction requires it.
• Export control on workload images. “The scheduler refuses images whose manifests declare restricted scientific domains.” The scheduler committee becomes a content regulator; coalitions that do not carry the filter become de facto unavailable in regulated jurisdictions.
• Provider licensure. “Registration fee or certification attestation before a beacon provider’s card is accepted.” A certification body emerges adjacent to the scheduler; entry is gated.
• Settlement as money-transmission. “The module verifies settlement clears a jurisdictional floor before clearing the grant.” The module is now a transaction filter; the operator is a money-transmitter by proxy.
• Reputation defamation or antitrust. “The coalition pins a reputation floor; non-compliant providers stop receiving grants.” A de facto ban; the provider’s recourse is litigation against the coalition operator.
• Platform liability. “The bridge retains ComputeLogs for N years under subpoena-friendly retention.” The log stream becomes a compliance artefact; the substrate is a records system.
• Jurisdiction over the scheduler committee. “Committee members must all reside in one jurisdiction for serviceability.” Committee geography becomes a legal foothold; the voluntary-members property slips.
• Continuity-of-service. “Retirement markers require N-day notice; early retirement triggers operator liability.” The operator’s exit guarantee (see web2 sustainability — operator exit) degrades from unconditional to gated.
• Gambling and fair-draw frameworks. “Randomness logs every draw’s use case to satisfy audits.” The substrate acquires an audit surface; beacon consumers become auditable subjects.

Each item is implementable as a content-field change to a module’s OrganismConfig or a TicketValidator clause. The slope is gentle. The destination is recognisable only after several steps.

How creep shows up in a proposal

A proposal that drifts into the deferred-primitives space usually does not name itself that way. It will read like a local optimisation with one of a few familiar framings. A temporary safeguard. Just a small content parameter. Optional by default. For the benefit of users. Because the regulator requires it.

None of the framings is wrong on its own. A temporary safeguard can be a reasonable ACL. A content-parameter change is how modules are configured in the first place. An opt-in flag empowers operators. User protection is obviously a goal. Regulators are real.

The question is where the filter is applied. An opt-in flag on an operator’s own TicketValidator is inside the blueprint. The same filter pinned in a module’s content, propagating to every provider on the substrate, is outside it.

How to tell

Four questions to run a proposal through. Failing any one means the proposal belongs on its own branch, with its own trust composition and its own audience.

1. Is the filter opt-in at a TicketValidator, or protocol-level in module content?
2. Who decides admission — the coalition operator, or the substrate?
3. Is the sanction voluntary degradation (a reputation score drops, subscribers unbond, a validator refuses new tickets) or compulsion (a module refuses to commit a properly-formed request because of who the requester is)?
4. Does the change re-introduce a supra-agent authority — a single party whose decisions the substrate is arranged to enforce?

A proposal that passes all four is a new TicketValidator clause, a new reputation organism, or a new module content field whose semantics are observational, not exclusionary. A proposal that fails any of them is a proposal for the deferred-primitives space. The discipline is clear: it lives on its own branch, rebuilds the trust composition from first principles, and earns its way.

What the stance provides

The stance this page resists is not “regulation is illegitimate”. It is “regulation is inevitable, so bake it into the substrate”. The blueprint rejects the second claim on engineering grounds. Buterin’s d/acc: one year later (2025) makes the closest kin argument for TEEs, zkproofs, and voluntary coordination rails as alternatives to substrate-level enforcement; the Legitimacy essay (2021) is the earlier version of why substrate-level enforcement forfeits the property that makes the infrastructure useful. Buterin’s On balance-of-power as a goal for the 21st century (2025) is the corresponding positive frame: once any sector’s natural limits recede, the remaining check is structural, and substrate-level enforcement removes a check rather than adding one. The deferred-primitives space is named because that is where substrate-level enforcement would land if it landed anywhere; the book defers it explicitly so the structural check stays.

The reasons here:

• Operator-boundary compliance composes. An operator facing a jurisdictional obligation adds a TicketValidator clause that requires whatever the obligation calls for — KYC evidence, an export-control attestation, a money-transmitter licence root. Every bonded member sees the clause, admits it or does not, and the composition stays readable.
• Substrate-level compliance does not. A filter pinned in module content propagates to every provider and requester, regardless of whether their jurisdiction would have imposed it. The module becomes a global regulator for a local concern.
• The operator’s exit is the substrate’s ultimate non-compulsion guarantee. A coalition operator facing unacceptable pressure retires the coalition; providers exit; requesters rebind elsewhere. Retirement markers (see sustainability) are the substrate’s answer to pressure on a named operator, not a substitute for the operator’s compliance posture.

Randomness

Randomness has the same structural exposure as Compute. Three quirks worth naming:

• Beacon providers in gambling or lottery jurisdictions are already licensed; a provider card admitting only licensed providers is the same admission slippage as Compute’s.
• Some jurisdictions require verifiable draws for regulated use cases (public lotteries, securities allocations). The module produces verifiable draws already; the slippage is when the module also enforces an audit-trail retention requirement.
• Beacon challenges by design go on Chronicle, not a module-owned stream. A proposal to move them into a dedicated compliance record — signed committee verdicts retained indefinitely for regulator consumption — is the same substrate-becomes-records-system pattern as Compute’s ComputeLog retention.

The four questions above apply identically.

Cross-references

• Topology-intro — deferred-primitives space
• Basic services — Compute
• Basic services — Randomness
• Threat model — Compute trust
• Part II — web2 overview
• Web2 sustainability — operator exit
• Buterin, d/acc: one year later (2025) and Legitimacy (2021).