Isotope Residual Audit Workflow: A Covariance-Aware Pipeline for Multi-Isotope Data

Overview This repository contains the open-source software, auditable data, and automated pipeline necessary to reproduce the informatics workflow and figures presented in the manuscript: "A Reproducible Covariance-Aware Workflow for Mass-Fractionation Residuals in Multi-Isotope Data". Rather than a simple calculation script, this package provides a rigorous covariance-aware decision architecture. It ingests multi-isotope data, linearizes it into logarithmic delta-prime coordinates, calculates mass-law residuals (Δ′), and algorithmically evaluates the completeness of the analytical metadata to determine the mathematically admissible output state. Admissibility Logic & Result States The core computational contribution of this software is its explicit validation of analytical covariance. Depending on the metadata supplied, the software enforces one of three output states: Formal (Diso): Executed only when a valid, well-conditioned ratio covariance matrix is supplied under a prespecified null model. Conditional Zero-Covariance (Diso,0): A sensitivity fallback executed only when full covariance is missing AND the user explicitly stipulates a zero-covariance assumption. Insufficient Metadata: The default safe state when necessary metadata is missing and no user stipulation is provided. End-to-End Demonstration (IM1 Spherules) As a reproducible feasibility demonstration, this package processes published Fe-isotope data from the CNEOS 2014-01-08 (IM1) spherules (transcribed from Loeb et al., 2023, arXiv:2308.15623, Table 6). Because the source material omits ratio covariance and external calibration metadata, the software correctly intercepts the dataset, denies formal statistical status, and explicitly limits the output to conditional zero-covariance sensitivity calculations. Repository Contents code/: The master Python execution pipeline (master_pipeline_all.py) that validates inputs, performs transformations, and renders figures. data/raw/: Immutable, manually transcribed source data and provenance ledgers. data/processed/: Auto-generated delta-prime coordinates and conditional mass-law residuals. output/: Machine-readable audit logs (status_report.json) proving the software's admissibility decisions. figures/: Output manuscript visualizations.