Isolate the contributions of narrative-interface changes

Determine the individual contributions of typed extraction, clearer instructions, reusable arithmetic receipts, and related changes in the revised narrative interface to its observed execution improvements, given that the changes were introduced together and the correction mechanism was not used.

Background

The paper compares an initial narrative interface with a revised interface that combines typed extraction, clearer instructions, and reusable arithmetic receipts. The revised interface achieves complete performance on the retained cases, whereas the initial interface exhibits failures involving calculator-call budgets, response formatting, missing calculations, and unsupported source attribution.

Because the interface modifications were introduced jointly, the reported results do not identify which modification produced which improvement. The absence of observed use of the correction mechanism further prevents assessment of its independent contribution. A controlled comparison that varies these components separately would be needed to determine their effects.