Building methods, frameworks, and software for validating causal mechanisms across the sciences.
Methods. Empirical methods for mechanism transport and evaluation validity: predicting when a finding transfers to a new system, and testing whether a benchmark measures what it claims — geometric invariants, matched controls, analytic failure bounds, and confound-first experimental design. Research Program →
Frameworks. Domain-general standards for validating mechanistic claims: establishing what the evidence warrants, resolving disputes about underlying mechanisms, and assessing transfer between systems. Mechanistic Validity →
Software. All code and data are public, results are preregistered, and boundary conditions are reported alongside successes. Open-source command-line tooling and Claude Code integration standardizes an auditable workflow. Reproducible Science →
People
Elliot Tower
Background in mathematics, philosophy, and machine learning. M.S. Computer Science (Data Science), UMass Amherst.