EmpiriForge

Selects, audits, and explains causal machine learning workflows for heterogeneous treatment effects and graph-based causal ML. Use for CATE, ITE, uplift modeling, S/T/X/R/DR learners, causal forests, matching, propensity scores, IV/DRIV, CEVAE/DragonNet, causal GNNs, graph neural networks with causal claims, therapeutic perturbation prediction, optimal intervention design, causal disentangled graphs, LLM-enhanced GNN mechanism identification, fault-diagnosis causal subgraphs, treatment targeting, policy personalization, Python CausalML/EconML-style projects, validation of causal-ML claims, and boundary checks between treatment-effect estimation and causal-invariant/stable prediction.

Orchestrates bilingual empirical research workflows from research question to reproducible paper package. Use when writing, reviewing, replicating, or packaging empirical economics/social-science papers; designing DID, IV, RDD, event-study, panel, synthetic-control, matching, or causal ML workflows; auditing tables and identification claims; checking method-source alignment; or turning research practice into reusable agent skills.

Orchestrates academic paper production from topic/proposal to manuscript, analysis, figures, submission files, and defense slides. Use for research-proposal planning, paper architecture, section-by-section drafting, evidence-gap tracking, statistical handoff, figure handoff, DOCX/PDF/LaTeX/PPTX delivery, revision workflows, and coordinating multiple research-writing skills without losing the paper ledger.

Builds and audits tabular-data prediction-model workflows for research papers. Use for clinical, environmental, biological, social-science, environmental-economics, or economics prediction models; tidymodels, random forest, XGBoost, logistic-regression baselines, partial least squares regression, PLS VIP, NDVI or environmental indicator models, enterprise carbon-emission forecasting, stable time-series prediction, causal-invariant prediction, distribution shift, cross-region/cross-industry/cross-policy validation, train/test splits, cross-validation, hyperparameter tuning, ROC/AUC, calibration, decision-curve analysis, bootstrap uncertainty, confidence intervals, prediction bands, variable importance, leakage checks, and manuscript-ready methods/results language.

Creates and audits publication-ready scientific figures in R. Use for ggplot2, corrplot, journal figure standards, TIFF/EPS/PDF export, COSTAR prompts, scatterplots, boxplots, bar charts, line plots, heatmaps, correlation matrices, correlograms, corrplot significance plots, volcano plots, Kaplan-Meier curves, forest plots, multi-panel figures, clinical figures, omics figures, ecology/environment figures, and reviewer-proof visualization checks.

Acts as a skeptical applied-economics identification reviewer. Use when evaluating DID, IV, RDD, event-study, panel fixed effects, synthetic control, matching, or causal ML designs; auditing robustness checks; preparing referee-style critiques; assigning causal-credibility verdicts; or strengthening the identification section of an empirical economics paper.

Work with Stata-based empirical economics projects. Use when reading, editing, running, or auditing Stata .do/.ado files, .log files, Stata replication packages, regression-table workflows, or Stata-to-R/Python conversion in applied economics. Can execute Stata when a local Stata executable is available; otherwise performs code/log/package review.