Vibe Coding in Software Development: A Multivocal Literature Review

Dataset description Replication package of the paper "Vibe Coding in Software Development: A Multivocal Literature Review". Authors: Shahbaz Siddeeq, Muhammad Waseem, Kai-Kristian Kemell, Mika Saari, Jussi Rasku, and Pekka Abrahamsson (Tampere University). What the dataset is The paper reviews 266 sources on vibe coding, 76 peer-reviewed and 190 grey. The sources were found by two database searches and two grey searches, in October 2025 and in September 2026, and are divided by publication date into window 1 (January 2022 to October 2025, 113 sources) and window 2 (November 2025 to 9 September 2026, 153 sources). Each source is classed as direct (146), adjacent (93), or background (27) by its relation to vibe coding. The review reports 60 findings under eight research questions, 51 on what the sources describe and 9 on reported outcomes, each with an evidence-strength label. This deposit holds every record behind the counts, tables, figures, and labels of the paper, the decisions of both authors who screened, scored, and coded the sources, the reconciliation of their disagreements, and the scripts that regenerate the tables and figures from the records. Files in the deposit - Main_Data_2026.xlsx: one workbook with every record of the review, 288 sheets in groups A to S. The first sheet, Contents, lists every sheet with what it holds, its number of rows, the file it copies, and the manuscript table, figure, or section it feeds. The sheet names in Contents are links. A reader needs no other file. - Supplementary_tables_2026.xlsx: the supplementary workbook of the paper, sheets S1 to S7. The paper has no appendix and refers to these sheets by name. - source_files/: the record files the main workbook copies, the scripts that read and write them, the frozen instructions given to the raters, the logs, and the generated LaTeX tables and figures of the paper. - README.md: the folder map, where each part of the record is, which file each table and figure of the paper comes from, and the run order of the scripts. - MANIFEST.csv: path, size, and SHA-256 of every file. - LICENSE: CC BY 4.0. Groups of the main workbook - A, corpus and sources: the 266 sources with their classification fields and extraction cells, the removed sources with reasons, the publication window per source, the coded dataset with the codes per research question, the codebook, the quality and credibility scores, the bibliographic record of every source, the definitions of vibe coding given by the sources, and the three grey sources authored by members of the review team with how they were handled. - B, search records: the search strings, databases, dates, and counts of both database searches, every retrieved record, the grey searches per channel and per query, and the duplicates found within and across the searches. - C, screening and selection: the title and abstract screening, the full-text screening, the quality stage, the grey eligibility and content review, the confirmed publication dates of grey sources, and the re-check of the exclusions of the October 2025 search. - D, quality and credibility: the item scores of every peer-reviewed source on the eleven-item checklist and of every grey source on the fifteen-item checklist, with band or tier, and the decisions at the quality stage. - E, extraction and coding: the extraction and coding records of the sources added in 2026 and of the reinstated sources, with the override files applied after reconciliation. - F, outcome records of RQ4: direction, measurement type, and horizon for every source and outcome pair, with who decided and when, and the frozen copies before the second author's check. - G, safeguards and reporting mode: every evaluated safeguard instance and the reporting mode of every workflow and safeguard item. - H, dependency decisions: the candidate pairs of dependent sources, the decision and reason for each, and the links the labels use. - I, labels and results tables: the criterion values and the label of every finding, the per-question tables, the corpus profile, and the coverage per research question. - J, sensitivity subsets: the labels on the direct sources alone and on other subsets, and the threshold sensitivity. - K, reliability checks: every check with its sample size, agreement, Cohen's kappa, PABAK, and number of disagreements, with the sample manifests. - L and M, decision files per stage: the decisions of the first author and of the second author, as entered before reconciliation. - N, reconciliations: every disagreement with its resolution, the resolved value, who decided, and the date, and the reconciliation worksheets. - O, the dataset of the October 2025 searches, as submitted: its eighteen sheets (summary, criteria, the four screening steps, the grey search, snowballing, the selected studies, quality and credibility scores, extraction, demography, codebook, code frequencies, thematic synthesis, evidence map, and data coding). - P, results tables of the submitted version. - Q, the tables and source lists of the paper as printed, parsed from the generated LaTeX. - R, documents: this description, the README and licence, the reliability protocol, the frozen instructions and rules, the calibration log, the check-by-check account of the reliability checks, the search guides, the codebook, the notes of the working workbooks, and the JSON records of the dependency and reporting-mode decisions and of the bibliographic data. - S, figures: the generated figures of the paper as images. Identifiers The record files identify each source by its record identifier (column `Corpus_ID`, for example WL74 or GL120). The paper numbers its source lists afresh, window 1 first. The sheet "Source ID to paper number" (group A, file `source_id_mapping.csv`) gives the paper number of every record identifier. The supplementary workbook uses the paper numbers. How to use the deposit Open Main_Data_2026.xlsx and start from the Contents sheet. To regenerate the tables and figures, use Python 3.11 or later with `openpyxl` and `matplotlib` and follow the run order in `README.md` from inside `source_files/`. The scripts read the record files and write the tables into `source_files/analysis/out_2026/`, the LaTeX tables and figures into `source_files/manuscript_tables_figures/`, and the main workbook into the top level. Versions The first version of this record (July 2026) held the dataset workbook of the manuscript as submitted, that is, the October 2025 searches. This version adds the records of the September 2026 searches, the revised classification and coding, the reliability checks and reconciliations, and the scripts, and keeps the earlier workbook under `source_files/` and as group O of the main workbook. Licence and citation The deposit is licensed under CC BY 4.0. Please cite the paper and this record.