Supplementary Data and Code for: Cross-Laboratory Generalisation and Conformal Uncertainty for CO₂–Brine Interfacial Tension: Closed-Form MARS Equations for Geological Carbon Storage Deployment

# IFT-SafePredict Supplementary Data & Code Archive This repository contains the dataset, pipeline scripts, mathematical models, and uncertainty quantification bounds accompanying the manuscript: **"Closed-Form MARS Equations with Calibrated Conformal Uncertainty for CO₂–Brine Interfacial Tension Prediction in Geological Carbon Storage"** (Olagunju et al., 2026). ## Directory Structure * **`data/`**: Contains the full 3,265-point master dataset separated by subcritical and supercritical thermodynamic regimes. Also contains the SHA-256 hashes of the locked External Validation (EV) sets to guarantee zero data leakage during training, and the hold-out test sets. * **`scripts/`**: Contains the full, reproducible Python pipeline. Run `run_pipeline_global.py` to sequentially execute data cleaning, regime classification, EV holdout, model training, and conformal interval calibration. * **`models/`**: Contains the fully trained, final Multivariate Adaptive Regression Splines (MARS) models (`.rds` format) for both regimes. * **`equations/`**: Contains the explicit closed-form algebraic equations (`.tex` format) extracted from the MARS models, alongside the Min-Max scaler parameters (`.json`) defining the orthogonal applicability domain bounds. * **`conformal_uq/`**: Contains the prediction interval half-widths calibrated on the hold-out test sets at 80%, 90%, and 95% nominal levels, plus summary reports of empirical coverage. ## Live Deployment A compiled, interactive web application embodying these models is publicly accessible at: [https://todak2000.github.io/ift-safepredict/](https://todak2000.github.io/ift-safepredict/)