Auditing SOP Compliance in CRM-Driven Remote Service Operations: A Human-in-the-Loop Framework with an Evidence-Sufficiency Discipline (proof-of-concept, synthetic evaluation)

Research paper and full reproducibility bundle for a human-in-the-loop framework that audits standard-operating-procedure compliance in CRM-driven remote service operations. The framework's distinguishing device is an evidence-sufficiency precondition on every control rule: when a record is missing, incomplete, delayed or ambiguous, the engine returns an indeterminate outcome rather than a compliance violation, so that 'the procedure was not followed' is never confused with 'the procedure was followed and not recorded'. The machine raises flags, an authorised human confirms violations, and every decision is bound into an append-only, tamper-evident audit history. The evaluation is entirely synthetic (1,100 generated CRM service cases, 16 control rules, declared ground truth, an independently authored test oracle) and is a specification-fidelity exercise: it is not a measurement of operational effectiveness, involves no human participants, and uses no employer or client data. The implementation is Python 3.12 standard library only, with a 105-test suite and an independent evidence audit that recomputes the published metrics from the saved outputs. The bundle retains the initial evaluation and an earlier superseded run, because the design was amended after the first outputs were inspected.