What is decision literacy?

Decision literacy is the ability to understand how a decision was reached, who or what made it, on which evidence and within which limits, and to judge whether it was sound. With AI agents, it means reading an agent's decision record and saying why the outcome was what it was, and what would have changed it.

Why does decision literacy matter with agents?

Agents produce many decisions, and people are asked to stand behind them. A customer complains that a refund was refused. A supervisor asks why a payment went through. A board member asks how a limit increase was approved. Someone has to answer, from the record, in plain words. Without decision literacy, the answer is "the system did it", which satisfies nobody and protects nobody.

It matters at review time too. A reviewer who understands why an action was held can judge the right fix. One who does not tends to approve or reject on instinct.

How does it work in practice?

In a governed platform, every action meets five bounds at once: the agent's role scope, its certificate, its sources, the customer's own limits and the platform guardrails. The lowest bound wins, and the verdict names which one decided. Every verdict is filed in an append-only record.

A decision-literate person reading that record can answer four questions:

1. What was the verdict: allowed, held, refused or withheld?
2. Which bound decided it?
3. What evidence was used, and how strong was its provenance?
4. What would have had to be different for another outcome?

The fourth question is the test. If you can answer it, you can explain the decision to a customer and challenge it if it was wrong.

What is an example?

A retailer's agent refuses a customer's request to change a delivery address after dispatch. The customer complains. The service lead opens the record: the verdict is refused, decided by role scope, because the agent's brief does not cover post-dispatch changes. The evidence was the dispatch status from the order system, sealed. The lead explains the reason, arranges the change by hand, and notes that the brief may need a held path for such requests. Retail and payments teams face this pattern constantly; see Article 4 literacy for agents in retail and in payments.

How is decision literacy different from data literacy?

Data literacy is about the evidence: where it came from, whether it is reliable. Decision literacy is about how evidence and limits combined into an outcome. A person may correctly judge that an invoice is external and still miss that the payment was actually held by the customer's own limit, not by the weak evidence. Both are needed to explain an outcome fully.

It is also distinct from understanding an automated decision as a category. Knowing a decision was automated tells you a person did not approve it; decision literacy tells you why it came out as it did.

Which terms are related?

• Decision chain: the five bounds every action passes through.
• Verdict: the outcome of the chain.
• Board literacy: directors need a version of this skill, covered in what AI literacy boards need.

For the full topic, read AI literacy and certification for people who work with agents.

Where the Academy fits

Pryme Intelligence Academy teaches the decision chain from the first track: Foundations covers why agents must be governed and the decision chain, and Operator includes how a governed agent decides and reading the workspace. Lessons are set on the real product, so learners read actual verdicts rather than diagrams.