Reproduction Artifact for Architectural Warning Coverage and Alert Burden in Microservices: A Retrospective Evaluation of Organizational Coupling

This artifact accompanies the manuscript “Architectural Warning Coverage and Alert Burden in Microservices: A Retrospective Evaluation of Organizational Coupling.” The study evaluates whether adding organizational coupling (OC) improves warning behavior beyond other predictors and examines the alert burden associated with that coverage. It uses twelve prepared two-year system histories comprising 1,260 system-week rows and 4,243 archived metric records. Matched models with and without OC, including separate history-adjusted comparisons, are evaluated against two supplied cohesion-related indicators: maxSSIC and stdLOMLC. The analyses distinguish coverage of high-indicator weeks from advance warning of eligible episode onsets. They report probability-score quality, alert counts, shared and separately selected threshold comparisons, simple reference policies, matched-count random timing references, and conditional paired repository-bootstrap intervals. Measurement audits document missing records, constant recorded indicators, and numerical sensitivity near outcome and alert thresholds. The package includes: The anonymous manuscript PDF and matching active LaTeX sources. Prepared input archives, frozen protocol specifications, and pinned Python dependencies. Complete experiment code, dated predictions, outcome labels, episode and alert ledgers, and per-system and aggregate results. Reproducible tables, numerical summaries, and native TikZ figures. Independent verification scripts, verification records, and a complete SHA-256 file manifest. A single entry point for regenerating and verifying the primary experiments and manuscript assets, with an optional PDF build. A fresh-extraction replay reproduced all 519 scientific CSV files identically and independently checked all 5,736 saved prediction scores. The rebuilt manuscript matched the reviewed PDF on every page. Verification scope and results are recorded in VERIFICATION.json and the verification directory. The primary evaluation retains a declared retrospective protocol. The results concern the supplied measurements and evaluated policies; they do not establish independently validated architectural deterioration, prospective warning performance, acceptable practitioner effort, or developer uptake. A separate temporal sensitivity analysis is explicitly identified and uses different outcome definitions and denominators. This review distribution omits identifying historical drafts and internal editorial material while preserving the scientific experiment files, ordinary scholarly references, and required third-party attribution. The README provides environment setup, reproduction commands, and interpretation guidance.