Declared source: what it means for AI agents

A declared source is a fact that a known person states but nothing independently proves: the income a customer types into a form, a note from a colleague saying a call took place, a supplier's answer on an onboarding questionnaire. An AI agent may use it to inform its work, but it carries less weight than a sealed or signed fact.

Where does "declared" sit among the provenance classes?

Every fact reaching an agent in a governed system carries one of four classes. From strongest to weakest:

1. Sealed: a record issued with a seal that can be checked against the issuer, so any change is detectable.
2. Signed: a fact a named, accountable party answers for, such as a policy approved by its owner or a ledger balance.
3. Declared: a statement from an identified, authenticated person, with nothing behind it but their word.
4. External: anything from outside the channels the business controls or has verified.

The guide to why source class matters explains how the four interact. The key feature of declared is that you know who said it. You do not know whether it is true.

Why does a declared fact need its own label?

Much of what an agent sees in a regulated business is declared. Customers describe their circumstances; staff summarise calls. Treating it as noise makes the agent useless; treating it as fact makes it dangerous.

The label lets the platform tell the two uses apart. A declared fact can shape a draft, a question or a triage decision. It should not, alone, be the evidence for an action that is hard to undo, and that rule is enforced at the moment of action.

How does an agent handle a declared fact in practice?

The agent records the fact with its class and its declarant, and the class travels with it into every draft. When an action depends on that fact and needs stronger evidence, the platform holds it and names the missing class in the verdict.

Take a small business lender. An applicant states annual revenue of 500,000 on the application form. The agent uses that figure to pre-fill the credit memo and to choose which documents to request. It cannot use the figure to approve the facility. The approval waits until bank statements arrive through a connected, authenticated feed, which turns the revenue figure into a signed fact, and a named credit officer reviews it.

How is a declared source different from a signed source?

Both name a party. The difference is accountability. A signed source carries the authority of a party you rely on for that kind of fact, such as the ledger for balances. A declared source is just a statement; the declarant could be mistaken, hopeful or lying.

A declared fact is further still from a sealed source, which can be checked against its issuer. An applicant's uploaded PDF payslip is declared. The same payslip issued with a verifiable seal by the payroll provider is sealed.

What should a provenance register record for declared facts?

Record who the declarant is, how they are identified, and which actions the fact may support. The source provenance register template has a column for each, and keeping it in one place supports one source of truth for what agents may do.

Where Pryme fits

The Pryme Context Protocol stamps every fact with its class as it reaches the agent, and a declared fact keeps that stamp through every draft. If an action needs a stronger class, the verdict is held or refused and the filed record names the declared fact that fell short. Read the governed agentic operating system, explained for how provenance sits alongside the other bounds.