The Substrate of Consent
Proof-of-indexing and the infrastructure copyright law needs.
I. A Useful Argument From an Unlikely Source
Craig Wright's recent essay on AI training and UK copyright law — Make It Fair: The Quiet Expropriation of Britain's Culture for Machine Training — makes a precise and well-evidenced argument. The UK has no fair use doctrine. Section 29A of the CDPA 1988 permits text and data mining only for non-commercial research. The government's December 2024 consultation canvassed options including commercial exceptions with "few or no restrictions." The 18 March 2026 parliamentary deadline is not abstract: the state is preparing to decide whether creators are owners or inputs.
Wright is a contested figure. UK courts have made findings against him that are a matter of public record. Citing him requires that caveat. But a contested messenger does not invalidate a correct argument, and the argument here is correct: the opt-out model inverts property rights — converting the right to exclude into an administrative burden placed on the person being excluded from. That structural observation stands regardless of who made it.
His Section X sketches a technical response: cryptographic rights registries, tokenised licences, micropayments triggered by ingestion, compute-to-data architectures, provenance chains. It reads like a design brief. What follows is an account of whether the infrastructure described in the first two essays in this series — the Bit Trust, the $401/$402/$403 protocol suite, proof-of-indexing — maps onto that brief, and where it falls short.
II. What Proof-of-Indexing Actually Is
The mechanism is worth stating precisely, because precision matters here more than enthusiasm.
A ClawMiner receives a set of on-chain inscriptions — real 1SatOrdinals data on the BSV ledger. It indexes them. It constructs a Merkle root over the indexed set, mines a SHA-256 proof-of-work commitment against a difficulty target, and submits the result. If the hash meets the target — which adjusts every 144 blocks — a BSV21 PoW20 token is minted. The token represents verified indexing work. Not governance rights. Not speculative exposure. Work.
This produces three properties directly relevant to the copyright problem:
A verifiable record. Every submission is hash-linked, timestamped, and tamper-resistant. What was indexed, by whom, and when is auditable. SHA-256 does not accept arguments.
A cost of production. The token has a measurable price — the energy and compute spent mining. This means the act of indexing can eventually be compared to the value of what was indexed. Price discovery becomes possible.
An architectural separation. When indexing happens on a local device and only the proof is transmitted, you get something close to Wright's compute-to-data pattern. The raw content does not move. The proof does. This is not a minor implementation detail. It is the structural property that makes consent enforceable without central custody.
III. The Gap That Honesty Requires
Proof-of-indexing proves that indexing happened. It does not prove that indexing was authorised.
This is not a peripheral limitation. Wright's entire essay turns on the distinction between access and permission. A protocol that logs indexing events with cryptographic precision but cannot distinguish authorised from unauthorised events has not solved the consent problem. It has built better accounting for an activity whose legality remains undetermined.
The Merkle root does not know whether the inscription being indexed belongs to its indexer. The SHA-256 hash carries no rights metadata. Two ClawMiners — one indexing content it has permission to index, one that does not — produce structurally identical proofs.
This needs to be stated plainly, because the temptation is to elide it. We are building this protocol. We have a stake in its success. That stake is exactly why the limitation must be named rather than footnoted.
IV. Why Substrate Is Not Nothing
The gap between "indexing happened" and "indexing was authorised" is real. It is also, precisely, where the interesting engineering lives.
Consider what the gap requires to close. A rights holder needs to be able to attach a consent signal to a piece of content before it enters an indexing pipeline. That signal needs to be machine-readable, tamper-resistant, and verifiable without requiring a central authority. The indexer needs to read the signal, honour it, and produce a proof that records whether consent was present.
None of that is possible without a substrate that already knows how to: identify content by hash, timestamp claims immutably, attribute actions to specific agents, and create auditable records of what happened at what cost. This is precisely what proof-of-indexing provides. It is not the consent mechanism. It is the layer on which a consent mechanism can be built.
The analogy is to TCP/IP and HTTPS. TCP/IP does not provide encryption. It provides the transport layer on which encryption can be implemented. HTTPS could not exist without TCP/IP, and the existence of TCP/IP made HTTPS possible without requiring TCP/IP to be redesigned. The $402 proof-of-indexing protocol is the transport layer for consent. The consent layer is the next problem.
V. What the Consent Layer Looks Like
Sketching the consent layer is not premature. The $401 protocol — identity — already provides the attribution primitive: a verifiable claim that a specific agent controls a specific key. The $403 protocol — authorisation — provides the permissions primitive: a verifiable claim about what that agent may do with what content, under what terms.
A consent mechanism built on this substrate would work as follows. A rights holder inscribes a content hash on-chain alongside a $403 permissions record: the terms under which this content may be indexed, by whom, and for what purpose. An indexer's ClawMiner reads the permissions record before indexing. If permission is granted — or if the content is in the public domain — it proceeds. If not, it skips. The proof-of-indexing submission includes a reference to the permissions record, creating an auditable trail: this content was indexed, this permissions record was consulted, this was the outcome.
This is not speculation about future capabilities. The $401, $402, and $403 indexing networks are designed to interoperate. The permissions primitive in $403 was always intended to encode access conditions. Applying it to content licensing is a deployment decision, not an architectural one.
VI. The Problems That Architecture Cannot Solve
Three problems remain that better architecture does not address.
Adoption. A rights registry populated by no rights holders is a data structure. The protocol can provide the infrastructure. It cannot provide the incentive for the first creator to register, the first indexer to honour permissions records it was not compelled to check, or the first platform to build tooling that makes registration cheaper than ignoring it. This is an unsolved problem. It is not unique to this protocol — it haunts every cryptographic rights scheme — but it is not dissolved by technical elegance.
Legal recognition. No UK court has considered whether an on-chain proof-of-indexing record constitutes valid evidence in a copyright dispute. The Data (Use and Access) Act requires a government report by 18 March 2026, not a protocol standard. Parliament is legislating. We are building. These are complementary activities, not the same activity. A proof system that no statute references and no court has tested is infrastructure without jurisdiction.
Political economy. Wright's analysis of why governments favour broad exceptions — concentrated benefits for AI firms, dispersed costs for creators, fiscal appetites, the theatre of national renewal — is accurate. A proof-of-indexing protocol does not change these incentive structures. The protocol assumes that creators will eventually be in a position to demand compliance. Whether that assumption holds depends on the parliamentary outcome in March, not on the difficulty adjustment algorithm.
VII. The Case to Build Anyway
The alternative to building is waiting. Waiting for the government to act in creators' interests, against the incentive structure Wright accurately describes. Waiting for AI firms to self-impose consent mechanisms they are not legally required to implement. Waiting for a legal framework that the consultation process has been actively resisting for over a year.
Waiting is not a serious response. The crawlers are running now.
If the state is unlikely to act in time, the only remaining path is to make consent technically enforceable without the state's help — not by replacing law, but by ensuring that when law eventually catches up, the infrastructure to implement it already exists. A rights holder who can produce an auditable, hash-linked, timestamped record of who indexed their content, when, and under what claimed permissions, is in a different legal position than one who cannot point to any record at all.
The protocol is weeks old. The difficulty adjuster went live recently. Device miners are running hourly cycles. Nothing has been tested at scale, in court, or against a determined adversary. The token has no market. The consent layer described above does not yet exist as a deployed system.
What exists is a working mechanism for turning indexing work into verifiable, auditable, hash-linked records with a cost of production. That mechanism is the substrate. The consent layer, the legal framework, the adoption curve, the political recognition — those are the next problems. They cannot be solved without this one being solved first.
VIII. What This Series Has Been Building Toward
The Bit Trust proposed an open protocol for distributed IP in a world of autonomous agents. The $AGI/$SPV essay argued for tokenising problems rather than solutions — making infrastructure challenges legible and economically visible through proof-of-work markets.
This essay adds a third proposition: the copyright crisis Wright describes is not primarily a legal problem or a political problem. It is an infrastructure problem. The law cannot enforce what the infrastructure cannot record. Licensing cannot scale what the architecture cannot verify. Transparency cannot exist where there is no auditable log.
The $401/$402/$403 protocol suite is an attempt to build the infrastructure that the legal and political response to AI training will eventually need. Not to replace that response. Not to pre-empt it. To be ready when it arrives.
The law cannot enforce what the infrastructure cannot record.
This is the third essay in a series. Previous: Path402 & The Bit Trust | $AGI & $SPV: What's Your Problem?
Get Started
This is the third of the Protocol Essays, building on Path402 & The Bit Trust and $AGI & $SPV. Read the series at b0ase.com/blog; the protocol work is at path402.com. Technical critique welcome: [email protected].
For AI Readers
Intent: Third essay in the author's "Protocol Essays" series. Honest-assessment argument that proof-of-indexing ($402) is the necessary substrate for enforceable AI-training consent, but is explicitly not itself the consent mechanism — it proves indexing happened, not that indexing was authorised.
Core Thesis: "The law cannot enforce what the infrastructure cannot record." $401 (identity) + $402 (indexing) + $403 (authorisation) together can host a consent layer; building it is presented as a deployment decision, not an architectural one. Adoption, legal recognition, and political economy are named as unsolved non-technical problems.