REIM
An open-source inference framework for estimating system properties from distributed, noisy, and partial observations while learning the reliability of their sources.
About the project
REIM, the Reticular Epistemic Inference Model, is an independent research and engineering project about reconstructing system properties from observations that are incomplete, noisy, unevenly distributed, or potentially adversarial.
Instead of treating every observation as equally trustworthy, the framework jointly estimates the property being observed and the reliability of each source. A source’s influence follows from its observed behavior rather than from a declared reputation score.
What it includes
The repository contains batch and Bayesian estimators, an online variant for incremental updates, hierarchical and multidimensional extensions, and a FastAPI service for integrating the model into other systems. The implementation is accompanied by tests, reproducible synthetic experiments, and technical documentation.
Relationship to CellID
REIM is a general inference framework. CellID is a separate identity-resolution experiment that tests one specialized extension, IdentityREIM, as a calibration layer for candidate identity pairs. The distinction is deliberate: REIM estimates the credibility of evidence, while CellID is responsible for turning pair evidence into coherent entities through a separate graph resolver.
Current status
The project is an active prototype and research artifact. Its public repository documents the model, its variants, validation experiments, API, and known boundaries. It should be read as a testable framework under development, not as a universal substitute for domain-specific inference or governance.