Evidence and sources

Research is only useful if it can be checked. Axe keeps the evidence for its work next to the work itself.

Every claim keeps its source

When a result states something it found, it records where: the document or page, the location within it, and the excerpt or value it relied on. Select a claim's numbered marker to see its source, and the sources list under the answer shows them all, so a reviewer can confirm a figure in one step.

Facts, calculations and assumptions

Results from an Agent go one step further and label each output by the kind of support behind it:

• Fact: taken directly from a source.
• Calculation: derived from other outputs. The result shows the formula and which inputs it used.
• Assumption: a judgement the method calls for, stated as one.

When evidence is missing

Axe does not fill gaps. When a value cannot be established, the result says so and gives the reason: not found, not applicable, access blocked, or unclear. The rest of the result still arrives, with the gap marked where the value would be.

Checks in a workflow

An Agent's workflow can include checks, steps that test the work against a rule in your method before it continues. Axe evaluates the rule in code against the values the run recorded, rather than asking the AI whether its own work is right. A run shows whether each check passed or failed, and which way the workflow went as a result. See Running an Agent.