Salmona: A Multi-Domain Architecture for Traceable Medical Billing Review in Colombian Healthcare

Abstract Background Medical billing audit in Colombia requires the simultaneous verification of administrative validity, clinical pertinence, and financial consistency of each claim, and may extend for more than three months per batch of submitted claims. Published artificial intelligence systems for this process focus on financial fraud detection through black-box classifiers, without covering the three audit domains simultaneously or describing concrete mechanisms of human governance. Methods A descriptive methodological study was conducted to document the design, architecture, and decision rationale of Salmona, a rule-grounded medical billing audit pipeline developed by Arkangel AI for healthcare reimbursement workflows in the Colombian health system. The description is based on the system’s technical documentation and the applicable Colombian regulatory framework. No patient data were collected and no empirical validation was conducted in this study. Results Salmona decomposes audit into three independent specialized audit layers: an administrative layer (27 rules, DAMA-UK instrument), a clinical pertinence layer (29 rules, PERT-CLIN instrument), and a financial and contractual layer (42 rules, FIN-CTR instrument). Each rule produces a structured result with three possible values: pass when there is evidence of compliance, fail when there is evidence of a violation, and n/a when the necessary information is absent. This innocent-until-proven-otherwise principle prevents the absence of a document from automatically becoming a billing objection. A canonical consolidation layer reconciles findings per invoice item, assigns objection causals according to Technical Annex 6 of Resolution 3047 of 2008, and calculates objected amounts under a financial integrity invariant. The human auditor reviews the consolidated result, may iteratively request corrections via comments, and initiates report delivery to the provider through an explicit action that is independent of the review cycle. Conclusions Salmona constitutes, to our knowledge, an early methodological contribution to the description of multi-domain medical billing review systems for the Colombian health system, with per-rule traceability and explicit human governance. Although initially oriented toward Colombian health insurers, the described architecture may also support provider-side pre-submission validation and structured response to reimbursement objections. The described architecture provides the basis for empirical validation with real claims in future studies.