Sheet NameDescriptionQuerySearch strings for all sourcesWhite_Literature_ProcessScreening log — white literature (peer‑reviewed)Grey_Literature_ProcessScreening log — grey literatureExtraction_FormData Extraction Form for the selected papersAccepted_Papers_WhiteGreyMaster inclusion table with full codingFinal_CodingSee detailed description belowOpen_EndedFree‑text purposes mapped to thematic codesPublication_VenueVenue aggregationLLMModel_TableModel‑usage summary (by family/model)Detailed Sheet DescriptionsQueryContains the exact search strings used in all databases and web sources. Each query is tailored to the syntax of the specific search engine (e.g., Scopus, IEEE Xplore, ACM DL, Web of Science, Google). Used to reproduce searches in the replication process.White_Literature_ProcessScreening log for peer‑reviewed literature. Includes staged inclusion columns:Title/Abstract (T/A) screeningFull‑text screeningSnowballing inclusionQuality Assessment (QA) Each stage records independent scores from three authors.Grey_Literature_ProcessScreening log for grey literature (e.g., blogs, technical reports). Mirrors the structure of the white literature sheet with the same staged decisions and per‑reviewer columns.Accepted_Papers_WhiteGreyConsolidated table of all included studies (white and grey literature). Contains complete metadata and coded variables such as:Work Category, Type of WorkArchitecture styles/patterns, Analysis Methods, Modelling LanguagesPurpose (Direction), Purpose (LLM), Human Model InteractionRationale, LLM Models, Model Enhancements, Prompt EngineeringUse Cases, Languages, and Future ChallengesOpen_EndedMaps free‑text responses to thematic codes for qualitative analysis of purposes and trends.Publication_VenueAggregates venues of included studies, including counts and venue type (e.g., conference, journal).LLMModel_TableSummarizes LLM families and models used in the included studies. Supports figures showing model adoption trends.CodingUsed for figure and table generation.Glossary of [White/Grey] Literature Process FieldsIncluded by T/A: Inclusion at title/abstract screening stage.Included by Fullread: Inclusion after full‑text review.Included by Snowballing: Added through citation chasing.Included by QA: Passed quality assessment stage.CitationIf you use this package, please cite the accompanying MLR paper.

Sheet NameDescriptionQuerySearch strings for all sourcesWhite_Literature_ProcessScreening log — white literature (peer‑reviewed)Grey_Literature_ProcessScreening log — grey literatureExtraction_FormData Extraction Form for the selected papersAccepted_Papers_WhiteGreyMaster inclusion table with full codingFinal_CodingSee detailed description belowOpen_EndedFree‑text purposes mapped to thematic codesPublication_VenueVenue aggregationLLMModel_TableModel‑usage summary (by family/model)Detailed Sheet DescriptionsQueryContains the exact search strings used in all databases and web sources. Each query is tailored to the syntax of the specific search engine (e.g., Scopus, IEEE Xplore, ACM DL, Web of Science, Google). Used to reproduce searches in the replication process.White_Literature_ProcessScreening log for peer‑reviewed literature. Includes staged inclusion columns:Title/Abstract (T/A) screeningFull‑text screeningSnowballing inclusionQuality Assessment (QA) Each stage records independent scores from three authors.Grey_Literature_ProcessScreening log for grey literature (e.g., blogs, technical reports). Mirrors the structure of the white literature sheet with the same staged decisions and per‑reviewer columns.Accepted_Papers_WhiteGreyConsolidated table of all included studies (white and grey literature). Contains complete metadata and coded variables such as:Work Category, Type of WorkArchitecture styles/patterns, Analysis Methods, Modelling LanguagesPurpose (Direction), Purpose (LLM), Human Model InteractionRationale, LLM Models, Model Enhancements, Prompt EngineeringUse Cases, Languages, and Future ChallengesOpen_EndedMaps free‑text responses to thematic codes for qualitative analysis of purposes and trends.Publication_VenueAggregates venues of included studies, including counts and venue type (e.g., conference, journal).LLMModel_TableSummarizes LLM families and models used in the included studies. Supports figures showing model adoption trends.CodingUsed for figure and table generation.Glossary of [White/Grey] Literature Process FieldsIncluded by T/A: Inclusion at title/abstract screening stage.Included by Fullread: Inclusion after full‑text review.Included by Snowballing: Added through citation chasing.Included by QA: Passed quality assessment stage.CitationIf you use this package, please cite the accompanying MLR paper.

No description available

No description available

This repository contains the replication package in which we provide the selected studies (white and gray) alongside the synthesized information.

LO2 datasetThis is the data repository for the LO2 dataset.Here is an overview of the contents.lo2-data.zipThis is the main dataset. This is the completely unedited output of our data collection process. Note that the uncompressed size is around 540 GB. For more information, see the paper and the data-appendix in this repository.lo2-sample.zipThis is a sample that contains the data used for preliminary analysis. It contains only service logs and the most relevant metrics for the first 100 runs. Furthermore, the metrics are combined on a run level to a single csv to make them easier to utilize. data-appendix.pdfThis document contains further details and stats about the full dataset. These include file size distributions, empty file analysis, log type analysis and the appearance of an unknown file.lo2-scripts.zipVarious scripts for processing the data to create the sample, to conduct the preliminary analysis and to create the statistics seen in the data-appendix.csv_generator.py, csv_merge*.py: These scripts create and combine the metrics into csv files. They need to be run in order. Merging runs to global is very memory intensive.findempty.py: Finds empty files in the folders. As some are expected to be empty, it also counts the unexpected ones. Used in data-appendix.loglead_lo2.py: Script for the preliminary analysis of the logs for error detection. Requires LogLead version 1.2.1.logstats.py: Counts log lines and their type. Used for creating the figure of number of lines per type and service.node_exporter_metrics.txt: Metric descriptions exported from Prometheus (text file).pca.py: The Principal Component Analysis script used for preliminary analysis.reduce_logs.py: Very important for fair analysis as in the beginning of the files there are some initialization rows that leak information regarding correctness.requirements.txt: Required Python libraries to run the scripts.sizedist.py: Creating distributions of file sizes per filename for the data-appendix.Version v3: Updated data appendix introduction, added another stage in the log analysis process in loglead_lo2.py

LO2 datasetThis is the data repository for the LO2 dataset.Here is an overview of the contents.lo2-data.zipThis is the main dataset. This is the completely unedited output of our data collection process. Note that the uncompressed size is around 540 GB. For more information, see the paper and the data-appendix in this repository.lo2-sample.zipThis is a sample that contains the data used for preliminary analysis. It contains only service logs and the most relevant metrics for the first 100 runs. Furthermore, the metrics are combined on a run level to a single csv to make them easier to utilize. data-appendix.pdfThis document contains further details and stats about the full dataset. These include file size distributions, empty file analysis, log type analysis and the appearance of an unknown file.lo2-scripts.zipVarious scripts for processing the data to create the sample, to conduct the preliminary analysis and to create the statistics seen in the data-appendix.csv_generator.py, csv_merge*.py: These scripts create and combine the metrics into csv files. They need to be run in order. Merging runs to global is very memory intensive.findempty.py: Finds empty files in the folders. As some are expected to be empty, it also counts the unexpected ones. Used in data-appendix.loglead_lo2.py: Script for the preliminary analysis of the logs for error detection. Requires LogLead version 1.2.1.logstats.py: Counts log lines and their type. Used for creating the figure of number of lines per type and service.node_exporter_metrics.txt: Metric descriptions exported from Prometheus (text file).pca.py: The Principal Component Analysis script used for preliminary analysis.reduce_logs.py: Very important for fair analysis as in the beginning of the files there are some initialization rows that leak information regarding correctness.requirements.txt: Required Python libraries to run the scripts.sizedist.py: Creating distributions of file sizes per filename for the data-appendix.Version v3: Updated data appendix introduction, added another stage in the log analysis process in loglead_lo2.py