How Do Humans Manage Conflicts Across Code and Models? A Controlled Scenario-based Study

This replication package accompanies a controlled scenario-based case study investigating how humans detect and resolve conflicts in Java source code and UML diagrams, and how consistent their behaviour is across these representations. Fifty-four student participants each assessed up to 44 scenarios: 20 matched Java–UML pairs addressing syntactic conflict situations, plus four additional semantic-contrast scenarios. The scenarios covered conflict, no-conflict, and pseudoconflict situations and were presented through 35 randomized form versions. Participants classified scenarios as conflict or no conflict, selected predefined resolution options or supplied free-text alternatives, explained their reasoning, and rated their confidence and perceived difficulty. The package includes the study protocol, scenario instruments, informed-consent and training materials, participant-coded form exports and processed response data, Python data-processing and statistical-analysis scripts, hand-maintained coding inputs, result tables and figures, explanations of the statistical methods, and detailed written answers to five research questions. These questions examine detection and resolution consistency, agreement across Java and UML, variation by structural cardinality and conflict kind, and associations with participant experience and scenario presentation position, including exploratory learning/fatigue analyses. Participant names have been replaced with participant codes, and the name-to-code mapping is not included. The shared records retain submission timestamps, background responses, and free-text answers. The package provides three entry-point scripts—scripts/build_dataset.py, scripts/build_extraction.py, and scripts/build_analysis.py—for rebuilding datasets and analysis outputs using the included form exports, coding inputs, and documented dependencies.