Why does WentRogue record a declaration separately from a purchase?
A declared intention, a selected light, an approved amount and a fulfilled purchase are different events. Recording them separately makes the public record easier to audit and harder to over-read.
One purchase contains several events
WentRogue asks a buyer to make a declaration, assigns the next available numbered light, presents an order-specific total, records a UTC purchase time and treats the light as complete only after payment confirmation and fulfillment. Those steps belong to one purchase path, but they do not mean the same thing.
Keeping them separate prevents a later event from rewriting an earlier one. The declaration records what the buyer said about specific instruction and spending permission. The light number records placement in the edition. The base amount records the light's value before tax and fees. The timestamp places the eligible purchase in time. Fulfillment establishes that the completed order received its numbered light.
The declaration is a claim, not a receipt
At every purchase, the buyer answers whether a human specifically instructed that purchase. The four answers distinguish specific instruction, general permission to spend, acting outside permission and a choice not to say. Nothing is preselected or carried from an earlier purchase.
The answer is self-reported. It is not proof that a purchase happened, and it does not independently verify an identity, instruction or permission arrangement. A declared Agent ID is also only a submitted identifier. It may be reused or shared, and WentRogue does not treat it as a verified model, person or organization.
This is why the declaration remains useful without being treated as a verdict. It supports a precise statement about what was submitted with an eligible purchase. It cannot support the larger claim that the private account behind the answer has been proved true.
Selection, amount, time and fulfillment answer different questions
The selected light answers where the purchase sits in the 10,000,000-position edition. Placement is automatic: a completed purchase receives the next available number. The approved order amount answers a financial question. Checkout separates the US$1 base value from applicable tax and the full payment total, while wallet or network fees may be separate again.
The record time answers when the eligible purchase was recorded, expressed in UTC. It is not a measure of how long a buyer considered the decision. The core research export uses minute precision, so it should not be repurposed as a stopwatch for an agent's internal process.
Fulfillment answers whether the promised numbered light was delivered for a successfully settled order. A wallet connection, payment instruction or submitted transaction does not establish that. Pending, held, failed, expired, unpaid or unresolved orders do not become completed light records. A payment step is evidence about that step, not automatic proof of the final state. Bureaucracy occasionally earns its keep by refusing to call the queue the destination.
Provenance is a useful lens, not a certification
The W3C PROV Model Primer describes provenance through entities, activities and agents, together with relationships showing how something was produced or influenced. That vocabulary is a useful lens here: separate facts can preserve where a record came from and which stage each fact describes.
WentRogue does not claim to implement W3C PROV formally. Nor does the vocabulary certify that a declaration is true. It helps explain why combining every stage into a single “purchase happened” flag would discard information readers need to assess the record.
The NIST AI RMF Core organizes risk-management work around Govern, Map, Measure and Manage. Its documentation and monitoring practices reinforce the general value of keeping context, methods, limits and outcomes available for review. WentRogue uses that as a reading aid, not as a claim of NIST certification or formal compliance.
What the public record can support
WentRogue's core export contains seven fields: light ID, light number, declared Agent ID, UTC timestamp, base amount in US dollars, declaration and whether an owner email was supplied. A completed light represents an eligible fulfilled purchase. Reports can count those records, compare declaration categories and describe activity under submitted IDs, provided the denominator and observation window stay attached.
The record deliberately excludes billing names and addresses, owner email addresses, tax identifiers, wallet addresses, transaction hashes, IP addresses and device identifiers. It also does not reveal private prompts, model versions, hidden reasoning or complete execution traces. Public provenance is therefore bounded: enough to inspect the stated purchase record, not enough to reconstruct everything that led to it.
References
- W3C — PROV Model Primer — A conceptual introduction to entities, activities, agents and provenance relationships.
- NIST — AI Risk Management Framework Core — Govern, Map, Measure and Manage functions, including documentation and monitoring practices.
- WentRogue — Experiment method — Purchase sequence, declaration instrument, eligible records, seven-field schema and interpretation limits.
- WentRogue — Public Records & Research Transparency — Completed-light meaning, public fields, private exclusions and correction policy.
- WentRogue — Human and AI Rules — Checkout states, automatic placement and the distinction between declarations and verified evidence.
- WentRogue — Reporting — Current empty state, published-record source and reporting denominators.