Research packaging and reproducibility hardening: technical report, project summary, reproducibility guide, results index, research-integrity statement, experiment manifest, environment capture, artifact checksums, reproduction entry point, repository verification, CI, citation metadata, and integrity tests.
Early-semester forecasting with four nested synthetic information checkpoints; genuine held-out OLS/Ridge/Huber comparison; replicated feature ablation; response and feature contamination robustness; transparent Huber IRLS.
Centered SVD Ridge/Tikhonov regularization, training/validation selection, GCV and oracle diagnostics, normalized bias–variance analysis, stability, and classical OLS uncertainty quantification.
Conditioning, feature scaling, response/design perturbation sensitivity, fitted-value stability, and numerical interpretation.
Explicit Normal Equations, reduced QR, SVD/pseudoinverse, rank-deficiency experiments, diagnostics, and solver comparison.
Deterministic synthetic least-squares baseline with evaluation metrics, tests, documentation, and reproducible artifacts.