Abstract Realist evaluations (RE) are increasingly popular in assessing health programmes in low- and middle-income countries (LMICs). This article reflects on processes of gleaning, developing, testing, consolidating and refining two programme theories (PTs) from a longitudinal mixed-methods RE of a national maternal and child health programme in Nigeria. The two PTs, facility security and patient–provider trust, represent complex and diverse issues: trust is all encompassing although less tangible, while security is more visible. Neither PT was explicit in the original programme design but emerged from the data and was supported by substantive theories. For security, we used theories of fear of crime, which perceive security as progressing from structural, political and socio-economic factors. Some facilities with the support of communities erected fences, improved lighting and employed guards, which altogether contributed to reduced fear of crime from staff and patients and improved provision and uptake of health care. The social theories for the trust PT were progressively selected to disentangle trust-related micro, meso and macro factors from the deployment and training of staff and conditional cash transfers to women for service uptake. We used taxonomies of trust factors such as safety, benevolent concerns and capability. We used social capital theory to interpret the sustainability of ‘residual’ trust after the funding for the programme ceased. Our overarching lesson is that REs are important though time-consuming ways of generating context-specific implications for policy and practice within ever-changing contexts of health systems in LMICs. It is important to ensure that PTs are ‘pitched at the right level’ of abstraction. The resource-constrained context of LMICs with insufficient documentation poses challenges for the timely convergence of nuggets of evidence to inform PTs. A retroductive approach to REs requires iterative data collection and analysis against the literature, which require continuity, coherence and shared understanding of the analytical processes within collaborative REs.