R

reproducible-research

Reproducible causal inference and econometrics in Python: simulations, quasi-experiments, experiments, causal ML and marketing mix models, with explicit identification assumptions and diagnostics.

Reproducible analysis of data-scientist wage dispersion and task allocation using BLS wage percentiles, sensitivity simulations, mixture models, and the Anaconda 2023 survey.

Python research code for segment-based change point detection, evaluated against human annotations on the Turing Change Point Dataset (TCPD). Constant, linear, mean-and-variance and Bayesian segment models with exact, pruned, greedy and online search are scored by covering and F1, the benchmark's published outputs are replicated, and every number in the accompanying article is checked against the committed results.

Python research code that measures false-positive rates and minimum detectable lift of geo-experiment estimators by running placebo experiments on real county sales data with no intervention. Compares difference-in-differences, synthetic control, ridge-augmented SC, conformal tests and structural time series.

Python research toolkit for measuring lateness accumulation, carried delay and recovery within train journeys in Great Britain, with validated event models and synthetic tests.

Python pipeline for estimating average and marginal CO2 emission factors in Portugal and Spain from ENTSO-E data, with econometric models, load-shifting experiments, and synthetic dispatch simulations.

Python research code for synthetic control, donor-pool geometry, and placebo inference. Includes DiD, ridge-augmented SC, synthetic DiD, Monte Carlo experiments, and the California Proposition 99 case study.

Python research code comparing interrupted time series (segmented regression) and Bayesian structural time series counterfactuals. Includes a Gibbs-sampled BSTS, Monte Carlo experiments on trends, seasonality, effect shape and triggered launches, and a placebo study of the 1983 UK seat belt law.

Python research code testing Prophet as a forecasting model. Includes random-walk and autocorrelation simulations, a hyperparameter tuning sweep, M4 hourly, weekly and daily benchmarks against naive, ETS, Theta and MSTL, and a rolling-origin evaluation on the Peyton Manning series.

R experiments behind the article "A Seed Is Not Reproducibility". The seed stays fixed while R's sampling rules, an extra random call, the locale, parallel workers, floating-point order or the data split change.

Python research code for the statistical power of text watermark detection. A token-level simulation of textGrain, green-list and Gumbel-max watermarks measures how detectability depends on passage length, entropy and edits, and tests what the published detection rates do and do not pin down.

Reproducible Python experiments turning treatment-effect estimates into targeting policies, with cross-fitted policy evaluation, uncertainty analysis, and policy trees on Hillstrom's randomised email experiment.

Reproducible Python analysis of heart-rate distributions across people, time, and activity states using NHANES, PhysioNet, and Fitbit data, with Box-Cox modelling, survey-weighted inference, and publication figures.

Reproducible EU plastics own-resource accounting, Portugal scenarios, and publication figures for The Plastic Tax That Plastic Does Not Pay. Includes source provenance, validated calculations, and a locked Python environment.

Reproducible Python experiments on ergodicity, mixing, and chaos: logistic and standard maps, Lorenz dynamics, and multiplicative growth. Code, five publication figures, numerical checks, and a locked environment accompany a separately published Medium article.

Python research code for sensitivity analysis in double machine learning. A twin design, in which the hidden confounder copies an observed benchmark covariate, tests a benchmark-calibrated omitted-variable-bias bound when the benchmark is itself estimated, with an application to 401(k) eligibility and a mathematical supplement.

Reproducible research code for the SRFM paper: special-relativistic geometry on OHLCV bars to split market regimes into timelike and spacelike.

Deterministic bounded search over inspectable signal-processing morphologies, with held-out evaluation.

Profile and selected evidence-first projects.

Open avian research methods protocols, schemas, reproducible analysis workflows, and synthetic demonstration resources for avian health, conservation breeding, and disease surveillance at Sunshine BirdWorld.