Projects
Interactive tools and open-source software — the same kind of decision tools I build for clients.
The applications below are working products of my research program. Each takes a statistical model and turns it into something a non-statistician can use — which is most of what good applied statistics comes down to.
Cricket
T20I win-probability forecaster
A live forecasting tool for Twenty20 International cricket. During a live match the chart below updates as the game unfolds; between games, pick any recent international from the menu to replay how the odds swung. Green is the side batting first, amber the chasing side, and the dashed line marks the innings break.
Live during matches; recent internationals otherwise. Win probability is for the team batting first.
The full interactive app — set any match state and explore the forecast — is deployed on shinyapps.io:
Tennis
Match win-probability tool
An interactive forecaster for professional tennis: choose any two players from the ATP or WTA tours, set the format and surface, and it returns a match win probability with the full distribution of likely set scores, plus a comparison against bookmaker odds. Currently in final development ahead of public deployment.
Soccer
International match forecaster
A win-probability and scoreline tool for international soccer built on tournament event data, including a match simulator with knockout and extra-time handling. In development, with a public release planned after the 2026 World Cup data becomes available.
Open source
Code on GitHub
dgt — an R package for reliability analysis. Implements generalizability-theory decision studies — the framework for answering “how many observations before this measurement is dependable?”
StatcastG — reliability at MLB scale. A Julia pipeline extending multi-facet generalizability theory to the massively unbalanced designs of real sports data, applied to MLB Statcast plate appearances (2015–2025).
github.com/rckaruna/StatcastG →
fairness-gtheory — how reliable is a fairness audit? R code for decomposing the uncertainty of AI fairness audits with multivariate generalizability theory — companion to the fairness-audit research under review.
github.com/rckaruna/fairness-gtheory →
MARE-code — designing the study before the data exist. Companion R code for Minimum Aberration for Random-Effect Designs: choosing the best generalizability-study design when measurement reliability is the goal.
github.com/rckaruna/MARE-code →
The models behind these tools are part of an active research program; methodological details appear in the associated publications as they complete peer review.