MRM Framework: Multi-Source Statistical Foundation for Canadian Carceral, Police, and Oversight Data

MRM provides a unified statistical foundation for analysing Canadian carceral, police, and oversight data by integrating five distinct data sources under a single set of estimators and identification primitives. The five sources span (i) the Ontario Tracking Information System (OTIS, A01-RCDD) provincial restrictive-confinement microdata, (ii) the federal Structured Intervention Unit Implementation Advisory Panel (IAP) reports together with independent academic replications, (iii) the Ontario Special Investigations Unit (SIU) police-oversight reports, (iv) the Toronto Police Service (TPS) open-data crime corpus and the Statistics Canada Crime Severity Index, and (v) the federal Corrections and Conditional Release Statistical Overview (CCRSO). The framework specifies a ten-estimator per-individual ensemble (inverse-probability weighting, augmented inverse-probability weighting, double-machine-learning interactive regression, propensity-score matching, partial-linear, and partial-linear super-learner variants), a generalised-estimating-equation clustering grid for high-cardinality groupings, a chi-squared family for aggregate contingency tables, a Mandela Rules classifier that operates at both the federal and provincial level, and a self-exciting Hawkes process for spatio-temporal modelling of TPS open-data events. MRM is implemented as a curated set of MRM modules in MORIE, a dual-language (R and Python) open-source toolkit released under GPL-2.0-only. Empirical replication of all published chi-squared statistics from the Sprott-Doob-Iftene academic record is demonstrated to within 0.01. A headline empirical finding: the provincial Ontario Mandela "torture"-classified proportion rises from 12.5% in fiscal 2023 to 20.6% in fiscal 2025, exceeding by 10.7 percentage points at peak the federal SIU rate of 9.9% documented in the published record. The companion R software paper is at DOI 10.5281/zenodo.20111233; the Python software paper is at DOI 10.5281/zenodo.20096350; the Hawkes methodology paper is at DOI 10.5281/zenodo.20102198.