Repository Analytics Engine v1
Initial extraction across five OSS systems; baseline 6-classifier benchmark; first FastAPI prototype.
First eight-paper draft series
Paper 1–Paper 8 outline. Identified that Paper 2 (dataset) and Paper 3 (classifier comparison) were thinly sliced and needed consolidation.
Self-healing locator framework
2,400 mutation events generated across seven refactor classes; tree-ensemble ranking models trained and calibrated. Paper D enters review.
RAITG & the test-generation service
312 requirements processed across banking, healthcare, e-commerce. 94.1% requirement coverage, 96.3% first-pass verify. Paper E completed.
file_age_days bug discovered & fixed
Sign bug from git log's default ordering contaminated ~40% of rows. Replaced with explicit min/max accumulators. All datasets and models rebuilt; commit_frequency emerges as a real secondary signal.
Eight papers consolidated to six
Paper 1–Paper 6 of the defect series merged into Paper A (empirical study), Paper B (cross-repo), and Paper C (risk-based testing). Total page count drops from ~80–100 to 20 with no loss of empirical content.
Production model gb-paper1-v4-fixed_age
Gradient Boosting + SMOTE deployed in the FastAPI risk-prediction service. GitHub Actions integration with advisory and gating rollouts.
Paper F — Post-Execution Defect Attribution
Vision paper proposing the fusion of dynamic test-failure signals with SHAP-explained repository priors for practitioner-actionable triage. Implementation roadmap drafted.
Telemetry-grounded follow-up to Paper C
Production-telemetry-based validation of risk-based prioritisation, the missing piece called out explicitly in Paper C's threats-to-validity section.