2026-10-13T01:55:05Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34002270 2026-09-26T05:25:52Z category_29194 item_type_3 month_year_09_2026

Reassessing Organizational Coupling for Architectural Early Warning in Microservices Anonymous authors (24423764) Software architecture Microservices pattern Reproducibility packageReassessing Organizational Coupling for Architectural Early Warning in MicroservicesThis release contains the paper's prepared source measurements, analysis code, dated predictions, complete results, protocols, tests, independent verification, and figure/table generators. <code>manuscript.pdf</code> is the current 11-page Overleaf compilation; <code>main.tex</code>, <code>sections/</code>, <code>bib.bib</code>, <code>Figures/</code>, and <code>tables/</code> are its matching sources. The scientific reproducibility boundary begins with prepared measurements, not upstream repository mining or independently validated architectural deterioration.Start with <code>ARTIFACT_MAP.md</code> for the exact paper-to-file map and <code>RELEASE_VALIDATION.json</code> for the completed fresh-extraction verification. <code>FILE_MANIFEST.json</code> records the SHA-256 and size of every other release file. The source archive is included, so experiments need no network after installing Python dependencies.<h2 dir="ltr">Install and verify</h2>Extract the complete ZIP, then open a terminal in the extracted directory containing this <code>README.md</code> and <code>reproduce.py</code>. Do not run individual files from inside a ZIP viewer. Use Python 3.12 or 3.13. Python 3.12.14 was used for this release check; the earlier follow-up was also reproduced on Python 3.13.13.Create the virtual environment outside the extracted package so that the strict manifest check does not see extra files. These commands invoke its interpreter directly; activation is unnecessary. In Windows PowerShell, with <code>python</code> pointing to Python 3.12 or 3.13:<pre>python -m venv ../oc-reproduction-env ../oc-reproduction-env/Scripts/python.exe -m pip install -r requirements.txt ../oc-reproduction-env/Scripts/python.exe -I -B verify_package.py ../oc-reproduction-env/Scripts/python.exe -I -B reproduce.py --quick --output-dir ../oc-quick-check ../oc-reproduction-env/Scripts/python.exe -I -B reproduce.py --full --output-dir ../oc-full-replay </pre>On Linux/macOS, with <code>python3</code> pointing to Python 3.12 or 3.13:<pre>python3 -m venv ../oc-reproduction-env ../oc-reproduction-env/bin/python -m pip install -r requirements.txt ../oc-reproduction-env/bin/python -I -B verify_package.py ../oc-reproduction-env/bin/python -I -B reproduce.py --quick --output-dir ../oc-quick-check ../oc-reproduction-env/bin/python -I -B reproduce.py --full --output-dir ../oc-full-replay </pre>Choose <code>--quick</code> for the short check or <code>--full</code> to recompute the experiments. The full replay includes the quick check's tests and asset checks, so running quick first is optional. The root <code>requirements.txt</code> installs the six pinned scientific libraries from <code>reanalysis/requirements.txt</code> plus pinned Pillow for image comparison; no separate dependency-install command is needed.Each output directory must be new or empty and outside the extracted package. To repeat a run, choose another output-directory name. Replays preserve the distributed expected results and place generated results, logs, and verification reports in that external directory. The entry point supplies its own import paths, so isolated Python execution does not depend on an author checkout or ambient <code>PYTHONPATH</code>.The quick check verifies file integrity, dependency pins, metric/provenance/temporal tests, and regenerates figure/table assets from saved results in a disposable copy. The full replay additionally rebuilds the retrospective analysis and measurement-presence audit from the included source ZIP, independently checks weekly aggregation and trace formulas, refits both temporal studies, checks complete results, and repeats parser sensitivity. Consult <code>python -I -B reproduce.py --help</code> for available options. Full verification includes independent calculations and sampled model refits; it is more extensive than comparing aggregate totals.Expect minutes for the full replay on a recent desktop, with substantial variation by CPU, filesystem, and environment. The release report records the actual environment, commands, durations, and outcomes rather than promising a cross-machine runtime. Dependencies and a TeX distribution are not bundled.<h2 dir="ltr">Find the reproduced results</h2>With the full command above, all paths in this table are relative to <code>../oc-full-replay/</code>. Quick-check outputs use <code>../oc-quick-check/</code> and contain the report, logs, and disposable asset copy; they do not contain newly fitted experiment results.<table><tr><th>Output</th><th>Contents</th></tr><tr><td><code>REPLAY_REPORT.json</code></td><td>Overall <code>status</code>, <code>expected_package_unchanged</code>, dependency versions, all stage outcomes, numerical comparisons, and input-bridge evidence. A successful replay reports <code>"status": "PASS"</code> and <code>"expected_package_unchanged": true</code>.</td></tr><tr><td><code>logs/</code></td><td>One log per executed test, experiment, generator, or independent check.</td></tr><tr><td><code>fresh_original/</code></td><td>Recomputed retrospective panels, dated predictions, ablation metrics, uncertainty, suppression, and sensitivity results; <code>_artifact/</code> contains the extracted public source.</td></tr><tr><td><code>fresh_measurements/</code></td><td>Recomputed record-presence mask, coverage, indicator diagnostics, and workbook schemas.</td></tr><tr><td><code>fresh_future/</code></td><td>Fresh base temporal predictions, labels, metrics, comparisons, and run manifest.</td></tr><tr><td><code>fresh_followup/</code></td><td>Fresh complete calibration/regularization grid, history comparisons, common-date results, and run manifest.</td></tr><tr><td><code>checks/</code></td><td>Independent retrospective, aggregation, trace-linkage, and temporal verification reports.</td></tr><tr><td><code>future_onset_summary/</code>, <code>future_onset_followup_summary/</code></td><td>Readable <code>RESULTS.md</code> and machine-readable <code>summary.json</code> for the fresh temporal outputs.</td></tr><tr><td><code>asset_and_parser_copy/Figures/</code>, <code>asset_and_parser_copy/tables/</code></td><td>Regenerated numerical figures and tables in a disposable package copy. Parser diagnostics are regenerated under that copy's <code>verification/future_onset_followup/</code>.</td></tr></table>The delivered <code>RELEASE_VALIDATION.json</code> records 48 passing tests, all 99 base-temporal CSVs and eight follow-up CSVs reproduced exactly, and the independent numerical checks. Your own <code>REPLAY_REPORT.json</code> records your run's outcome; a saved validation report does not substitute for executing the replay.<h2 dir="ltr">Included analysis code</h2>The release includes 39 Python source files: 25 directly in the extracted release and 14 in the unchanged nested public source archive. No analysis repository or additional private code is required to reproduce the reported prepared-data experiments.<table><tr><th>Code</th><th>Purpose</th></tr><tr><td><code>reproduce.py</code>, <code>verify_package.py</code></td><td>Complete replay entry point and strict file-integrity verification.</td></tr><tr><td><code>reanalysis/run_fast_oc.py</code>, <code>reanalysis/fast_postprocess.py</code></td><td>Retrospective ablation, timing references, suppression, sensitivity, and conditional uncertainty.</td></tr><tr><td><code>reanalysis/measurement_validity.py</code></td><td>Measurement-presence and prepared-indicator audit.</td></tr><tr><td><code>reanalysis/run_future_onset.py</code>, <code>reanalysis/run_future_onset_followup.py</code></td><td>Base temporal forecasts, matched history, calibration/regularization grid, and common-date comparisons.</td></tr><tr><td><code>reanalysis/test_metrics.py</code>, <code>reanalysis/test_future_onset.py</code>, <code>reanalysis/test_future_onset_followup.py</code>, <code>scripts/test_verify_reproduction.py</code></td><td>The four test modules, totaling 48 tests.</td></tr><tr><td><code>scripts/build_paper_assets.py</code>, <code>scripts/build_temporal_paper_assets.py</code></td><td>Numerical figure/table generators.</td></tr><tr><td><code>scripts/summarize_future_onset.py</code>, <code>scripts/summarize_future_onset_followup.py</code>, <code>scripts/verify_reproduction.py</code></td><td>Readable result summaries and original-result reconciliation.</td></tr><tr><td>Nine Python files under <code>verification/</code></td><td>Independent metric/oracle checks, bootstrap checks, workbook aggregation, trace linkage, follow-up package checking, and parser sensitivity. Exact paths appear in <code>ARTIFACT_MAP.md</code>.</td></tr><tr><td>Fourteen Python files inside <code>source/anonymous_oc_early_warning_artifact.zip</code></td><td>Original modeling and data helpers, including the complete <code>analysis/hazard_early_warning/hazard_case_study/</code> package imported by the retrospective runner. The replay extracts this archive automatically.</td></tr></table><h2 dir="ltr">Contents and analysis boundaries</h2><ul><li><code>source/anonymous_oc_early_warning_artifact.zip</code>: exact public prepared-data archive, including all 24 nested input ZIPs, archived modeling helpers, system manifest, and reconciliation outputs. See its source notice for attribution and license.</li><li><code>reanalysis/results/</code>: 12 prepared panels, dated retrospective sequences, ablations, timing references, conditional intervals, suppression results, and threshold/horizon sensitivity.</li><li><code>reanalysis/measurement_validity_results/</code>: workbook schemas, all 1,260 weekly record-presence observations, per-system coverage and indicator diagnostics.</li><li><code>reanalysis/future_onset_results/</code>: native temporal backtest, including every label/prediction inventory and unavailable case.</li><li><code>reanalysis/future_onset_followup_results/</code>: complete calibration/regularization grid, matched history comparisons, exact training-anchor metadata, and common-date analyses.</li><li><code>verification/</code>: independent numerical checkers, declared protocols, parser sensitivity, public trace-linkage checks, and frozen historical verification reports.</li><li><code>scripts/</code>, <code>Figures/</code>, <code>tables/</code>: reproducible manuscript assets and explicitly identified supplementary displays.</li></ul>RQ1 audits measurement presence. RQ2 compares temporally controlled OC forecasts with controls and prior indicator history. RQ3 separately diagnoses inherited retrospective scores and notification suppression. These branches have different labels, dates, units, and denominators. They cannot be compared as if they were the same prediction task.Headline checks: 839 recorded and 421 unrecorded system-weeks; primary native prepared/recorded Brier better/worse/tied counts 9/7/4 and 4/8/2; matched-history recorded counts 5/5/2 with only five eligible onsets. Diagnostic two-week suppression changes Full alerts from 301 to 128 while retaining 105 event weeks and 30 onsets. The plot retains a larger inventory, explicitly marked in its caption. See the paper and artifact map for denominators.<h2 dir="ltr">Regenerate the paper</h2>The replay regenerates numerical assets without changing this release. For manual exploration, work in a disposable copy and run:<pre>python scripts/build_paper_assets.py python scripts/build_temporal_paper_assets.py pdflatex -halt-on-error main.tex bibtex main pdflatex -halt-on-error main.tex pdflatex -halt-on-error main.tex </pre>A normal TeX installation needs IEEEtran, TikZ, AMS packages, booktabs, balance, microtype, hyperref and the standard packages named in <code>main.tex</code>. Tectonic is an alternative. No shell escape is required. The delivered PDF was checked using Overleaf's TeX Live 2026 build; pagination/fonts may vary across TeX installations. <code>Figures/tikz/standalone_diagrams.tex</code> is a separate four-diagram preview, including a supporting suppression schematic absent from the paper. 2026-09-26T05:25:52Z Dataset Dataset 10.6084/m9.figshare.34002270.v1 https://figshare.com/articles/dataset/Reassessing_Organizational_Coupling_for_Architectural_Early_Warning_in_Microservices/34002270 CC BY 4.0