Technical and Non-Technical Debt Across ISO/IEC 5338 Lifecycle Processes: A Joint Analysis in AI-Integrated Projects

Replication package for “Technical and Non-Technical Debt Across ISO/IEC 5338 Lifecycle Processes: A Joint Analysis in AI-Integrated Projects” This package contains the data, mapping rules, prompts and analysis code of the study. The study maps technical debt (TD) instances from industrial AI-integrated projects to the lifecycle processes of ISO/IEC 5338:2023, ISO/IEC/IEEE 12207:2017 and ISO/IEC/IEEE 15288, and to the process areas of the People Capability Maturity Model (People CMM). It then analyzes 14 technical and non-technical debt categories jointly, using a subset multiple correspondence analysis (MCA) at instance level and a Jaccard similarity network at project level. Files 1. The_Mapping_Tables_and_Prompts.xlsxThis Excel workbook contains the final mapping tables, the Large Language Model (LLM) False Positive (FP) and False Negative (FN) verification prompts used in the study, and the supporting data on the debt taxonomy and the interview dataset. It is organized into the following worksheets: ReadMe: Describes the workbook and all attributes of the mapping tables. TD Taxonomy: Shows the 14 main debt categories (six non-technical, eight technical) and their subcategories. NTD_Mapping: Contains the mapping of the instances of the six non-technical debt categories to ISO processes and People CMM process areas. TD_Mapping: Contains the mapping of the instances of the eight technical debt categories to ISO processes. FPV: Contains the specific prompt used for False Positive verification. FNV: Contains the specific prompt used for False Negative verification. Process_abbreviations: Contains the abbreviated and full names of all mapped processes. Interview_demographics: Shows the characteristics of the participants, projects and companies. 2. Mapping_Guideline.docxThe formal protocol used by the researchers to identify candidate processes, assess their relevance to a TD instance, and document the justification of each mapping decision. 3. Interview_guide.pdfThe semi-structured interview guide used to collect the dataset (English translation). 4. Subset_MCA.ipynbA Jupyter Notebook containing the Python implementation of the subset multiple correspondence analysis (MCA) with project weights (RQ2a), including the leave-one-project-out stability analysis (RV coefficients), the contributions and coordinates of all points, and the MCA maps. 5. Subset_mca.RAn R script containing the functions of the subset MCA: project weights, subset MCA, supplementary points, RV coefficient and leave-one-project-out analysis. 6. Subset_mca_analysis.RAn R script that reads the mapping tables from The_Mapping_Tables_and_Prompts.xlsx, prepares the combined data, and runs the subset MCA, the stability and permutation analyses, and the cross-check with the R package FactoMineR. It uses the functions in subset_mca.R; both files must be in the same folder. 7. MCA_results_R.zipThe output of subset_mca_analysis.R. It contains the following files: subset_mca_output.txt: Complete console output of the analysis. eigenvalues.csv: Eigenvalues and percentages of subset inertia of all dimensions. process_coordinates_contributions.csv: Weights, coordinates and contributions of the active processes. debt_category_coordinates.csv: Coordinates of the debt categories (supplementary). ml_domain_coordinates.csv: Coordinates of the ML domains (supplementary). supplementary_process_coordinates.csv: Coordinates of the processes mapped in fewer than three projects (supplementary). leave_one_project_out_rv.csv: RV coefficients obtained when each project is excluded. permutation_baseline.csv: Observed eigenvalues compared with those of 200 permutations. sessionInfo.txt: R and package versions used. 8. Jaccard_implementation.ipynbA Jupyter Notebook containing the Python implementation of the project-level Jaccard similarity network (RQ2b), including the exclusion rules and the network visualization. Reproducing the analyses The numerical results of the subset MCA (RQ2a) reported in the paper were computed in R with subset_mca_analysis.R, which uses the functions in subset_mca.R and reads the mapping tables directly from The_Mapping_Tables_and_Prompts.xlsx. To run it, both R files and the workbook should be placed in one folder. The output is written to a results folder, whose contents are provided in MCA_results_R.zip. Subset_MCA.ipynb implements the same subset MCA independently, reproduces the R results and creates the MCA maps shown in the paper. Jaccard_implementation.ipynb computes the project-level Jaccard similarity network (RQ2b) and creates the network figure shown in the paper. Both notebooks read the sheets NTD_Mapping and TD_Mapping as input. File paths in the first cell of each notebook should be set to the location of the workbook or of the exported sheets. Notes Anonymization: names of participants, companies and projects were removed. Projects are identified by codes, and the projects of one participant are merged into one project unit (e.g., P10.1–P10.3 into P10). Copyright: the columns of the mapping tables that quoted text from the ISO/IEC standards and the People CMM (process purposes, activities and tasks, and details) were removed for copyright reasons. Each mapping refers to the relevant clause, so readers with access to the standards can look up the text. Language: the interviews were conducted in Turkish; the case descriptions, root causes and the interview guide were translated into English. LLM use: GPT-4o mini supported the verification of the mapping, and every LLM output was reviewed by a researcher before any change was made to the mapping.