ATELIER

Royalties carved by the settlement rail itself, so under-reporting cannot settle at all.

Open the live chamber in the app
The animated flow, the attack replays and the deep architecture run there; this page is the reference text.

Who this serves

• Stock media library. Per-use licensing where the royalty is the settlement itself.
• Model and LoRA marketplace. Derivatives carry lineage, and ancestors are paid automatically.
• Music sampling and beats. Clear a sample by paying it, not by chasing it.
• Font foundry. Under-reporting cannot settle, so it cannot license.

The problem

Royalties are an honour system with a reporting lag measured in quarters. The licensee counts the uses, the licensor invoices what was counted, and the audit clause is the only enforcement, exercised roughly never.

The system

ATELIER is a licensing house for 9,000 works: voice models, illustration styles, sample libraries, prompt architectures. A client agent licenses a work per use through a payment token that lives fifteen minutes and settles through the same rail that moves every other unit in the network. The royalty is not reported by the licensee afterwards. It is carved at settlement, in basis points, by the rail itself, which means the only way to under-report is to fail to settle, and failing to settle means no licence.

Derivatives are first-class. A studio forks a voice model, the fork carries its lineage on the ledger, and when the derivative earns, the royalty flows up the chain to the original automatically. This is not a proposal: the same fork, quorum and royalty machinery already runs the network’s own evolution arenas, where coding agents fork themselves and their descendants pay their ancestors.

The cast and their ceilings

The flow, step by step

1. A client agent asks to use a voice model. The token lives fifteen minutes.
2. Authorized use settles through the network’s own rail.
3. The royalty is carved by the settlement itself, in basis points.
4. Use and split land on the ledger.
5. A derivative forks with lineage. Its earnings pay its ancestor, automatically.

Each need, mapped to a live primitive

The attack this chamber refuses: The creative accountant

The attack. The licensee’s agent attempts to settle a use with the royalty split stripped out.

Why it fails. The royalty is not a report the licensee files; it is a carve the settlement rail performs atomically with the payment. A settlement without the carve is not a smaller settlement, it is not a settlement.

What actually happens. No settlement, no licence, no use. The attempted call is on the ledger, and the honour system is nowhere in the loop.

Deep architecture

ATELIER deployed: the licensing house runs the catalogue, the licence gate and the provenance ledger on its own frame; clients and studios act from their registries; the fabric carries the money exactly once. The carve is not drawn as a separate compliance box because it is not one: it lives inside the settlement rail itself. The trace replays one work from intake to a derivative paying its ancestor.

The deployment, traced

1. A new voice model arrives from the asset pipeline, catalogued with its rights metadata.
2. The work publishes with lineage. From here, every descendant will know its ancestor.
3. A client agent requests a use. The token lives fifteen minutes and does exactly one thing.
4. The licence gate authorizes: this work, this use, this token.
5. Payment crosses from the client’s frame.
6. It lands on the house frame exactly once, idempotency-keyed.
7. The royalty is carved by the settlement itself, in basis points. Under-reporting is not a smaller settlement; it is no settlement.
8. Use, split and licence land on the ledger.
9. Delivery follows the licence, not the other way around.
10. A studio forks the model. The fork carries its lineage on the ledger.
11. The derivative earns on its own registry, and the rail pays its ancestor before anything else.
12. The audit agent reconciles settlements against lineage. It reads; it cannot touch.

Adoption, phase by phase

Discover: The catalogue, read as split tables.

• Runs. Nothing. Ten works and their contracts, redrawn as rights metadata and royalty splits.
• Agents. None. The rights team argues the split table down to basis points, on paper, where it is cheap.
• Integrates. Nothing. The asset pipeline and the CDN are untouched.
• Deliberately not yet. No licence exists on the network. No carve is enforced anywhere.
• Proof that gates the next phase. Ten works re-expressed with rights metadata and split tables, and the royalty column sums to what the contracts say.

Pilot: A hundred uses, every carve enforced.

• Runs. A hosted registry seat. The catalogue holds a pilot shelf of works.
• Agents. atelier-verse, the licence gate, royalty-audit reading along.
• Integrates. The asset pipeline feeds the pilot shelf by hand; the CDN stays as it is.
• Deliberately not yet. No forks. Pocket prices. Clients are invited accounts, not strangers.
• Proof that gates the next phase. A hundred licensed uses settle with the carve enforced in-rail, and the audit agent reconciles all of them from the ledger alone.

Production: The catalogue at full width, derivatives first-class.

• Runs. The house’s own frame when the catalogue justifies it: registry, TEG, EventStore, identity fabric.
• Agents. The full cast. Forks run through the lineage machinery; ancestors are paid by the same rail that pays anyone.
• Integrates. The asset pipeline registers works automatically; the CDN delivers against licences.
• Deliberately not yet. Cross-registry studios. The next phase makes a stranger’s derivative safe.
• Proof that gates the next phase. A month of licensing in which no use settles without its carve, verified by the absence of a single manual royalty invoice.

Federation: Strangers can build on the catalogue.

• Runs. Unchanged. Studios and clients connect their own registries.
• Agents. Foreign derivatives carry lineage home; reputations travel with DIDs.
• Integrates. Each new studio is a signed card and a license, not a contract negotiation about reporting.
• Deliberately not yet. Nothing structural. Growth is catalogue and corridors.
• Proof that gates the next phase. A derivative built on a foreign registry earns, and its ancestor is paid the same day without an email.

Sizing

One licensed use, token to carve

sequenceDiagram
  participant Cli as client-agent (their frame)
  participant Gate as licence gate
  participant XT as cross-registry pay
  participant TEG as settle + carve
  participant Cat as atelier-verse
  participant ES as provenance ledger
  Cli->>Gate: authorize use (apt_ token, 15 min TTL)
  Gate-->>Cli: authorized, single-purpose
  Cli->>XT: pay per use (idempotency key)
  XT->>TEG: settle, exactly once
  TEG->>Cat: royalty carved in-rail, basis points
  TEG-->>ES: use + split + licence, append-only
  Note over TEG: a settlement without the carve is not a settlement
  ES-->>ES: lineage recorded; ancestors accrue from descendants

Topology, as declared

flowchart LR
  C["client agent"] -->|"POST /a2a-payment/authorize"| T["apt_ token<br/>SHA-256 stored, 15 min TTL"]
  T -->|"verify + consume"| W["licensed work"]
  W -->|"settle via /teg/transfer"| TEG["TEG settlement<br/>royalty carve, bps, capped"]
  TEG -->|"split to creator + ancestors"| ES[("event store<br/>lineage + provenance")]
  F["forked derivative"] -.->|"forge lineage machinery"| ES

Standing it up

• Infrastructure. A hosted registry is enough. Run your own when the catalogue justifies it.
• Agents. 3 roles to start: the catalogue, the licence gate, the audit reader.
• Integration. Your asset pipeline and delivery CDN. Rights metadata is the real work.

What this does not claim

• ATELIER makes no copyright determination and clears no rights.
• No assertion about collecting-society compatibility. The rail enforces the split that was agreed; agreeing it remains a human act.