This research addresses the challenge of variation in the interpretation and application of safety risk judgements within ultra-safe systems. Despite the formalisation of safety management frameworks, safety risk decisions continue to rely on practitioner judgement. This may affect the consistency, reliability, and defensibility of safety management within high-hazard domains.

To address this problem, this research developed and initially validated the Learning Informed Safety Assessment (LISA) tool. This is an empirically derived decision support mechanism designed to improve the consistency and defensibility of safety-critical decisions. The tool should be understood as a bounded, cue-based model derived from empirical indicators under specified conditions, rather than a comprehensive representation of safety risk decision-making.

Study 1 used qualitative interviews across safety-critical domains to collect safety risk management scenarios, which were coded using an adapted Gioia methodology and analysed through clustering and similarity techniques. Study 2 transformed these qualitatively derived cues into a binary dataset, reducing narrative accounts into structured empirical representations to develop a fast and frugal decision tree using the FFTrees package in R.

The optimal decision tree was operationalised as a decision support tool incorporating four cues and two modifiers (“unless” conditions). The tool yielded 97% sensitivity, 80% specificity, and 89% balanced accuracy when tested on the internal dataset. When applied to 24 independent real-world cases, the tool correctly classified all cases within the bounded validation sample. These findings demonstrate proof-of-concept performance under the study conditions; however, additional empirical testing across larger and more diverse safety-critical contexts is required before broader practical application.

This research makes theoretical, methodological, and practical contributions. It advances understanding of how variation in safety risk judgement may affect safety management, demonstrates a replicable mixed-methods framework for decision-support tool development, and produces a structured, evidence-based mechanism intended to support safety-critical decision-making in domains such as nuclear energy and civil aviation.