A Source-Linked Review Method for KPI Definition--Implementation Disagreements

Abstract Enterprise key performance indicators (KPIs) can have different documented and executable calcu-lations even when software continues to return plausible values. Detecting such a mismatch does not establish which formula or boundary rule is authorized, which source evidence supports the finding, or whether every decision needed to complete review has been recorded. We present a source-linked review method that connects a declared KPI representation and an executable implementation to verified input-level comparisons, separately justified regional claims, individual review questions, and question-specific disposition evidence. The method includes an executable validator for evidence identities, source locations, comparison findings, approval content, and derived review status. Two preserved Literate Programming for Automated KPI Extraction (LEAP) demonstration cases were examined using exact rational reference calculations and direct execution of archived Python functions. Eight source probes produced four value differences, three definedness differences, and one agree-ment. A separate Ethylene boundary execution and a constructed joint-undefinedness control tested additional conditions. Five valid review records were accepted and all 26 fault-injected records were rejected. A syntactic inventory of seven criteria across two cases found 5/14 positive dossier checks and 14/14 structured-record checks. These results establish a reproducible comparison-and-authorization contract for the evaluated cases.