When both agents are sure.

The control room compares two claim strings after trimming outer whitespace and converting letters to lowercase. Different strings produce zero percent agreement. Replacing the second claim with an uppercase copy of the first, with spaces at the edges, produces one hundred percent. At three pixels per displayed percentage point the bars span zero and three hundred. This is binary string equality, not semantic similarity or confidence in a payment outcome. Two matching claims can both be false. The tool preserves the operator context for review rather than deciding which agent is truthful. Matching is not enough when both claims are empty. The source also requires the normalized first string to have positive length. A nonempty identical pair gives one hundred percent, while deleting both claims gives zero and Conflict detected. At three pixels per percentage point the bars span three hundred and zero. Adding punctuation to only one claim also breaks equality, even when a human would understand the same meaning. This strict test creates a clear handoff signal but does not understand evidence, resolve ambiguity, or justify an irreversible transaction. The initial panel displays two evidence references and provides two editable reference fields. If both fields are cleared, the number in the panel remains two because the update function does not recompute it. At one hundred pixels per reference the displayed-count bar spans two hundred while the populated-field-count bar is zero. The exported JSON contains the actual empty inputs alongside computed agreement and status. Saving stores the case in this browser, and export downloads a local handoff. Neither checks the referenced documents or contacts a payment provider. Inspect the fields rather than treating the static reference count as proof.