Research & Analysis Workflows

Audit an existing .bib file entry by entry against canonical metadata (Crossref, OpenAlex, arXiv, Zotero) to catch silent citation errors that pass spell-check and kill peer review, including wrong years, mis-cited authors, wrong journal or volume, swapped title-author pairs, dead DOIs, and hallucinated entries from LLM-assisted drafting. TRIGGER on "/bibcheck", "audit my .bib", "verify my references", "check my bibliography", "are any of these citations fake", "did Claude hallucinate a citation", "pre-submission bib audit", or before shipping a manuscript or R&R to Marketing Science, JMR, JCR, or Management Science. Read-only: writes a report and a separate corrected.bib, never touches the source.

Triage any causal question to the identification strategy the data can support, then hand off to the owning method skill; owns the selection-on-observables branch (overlap, doubly robust estimation, double ML, causal forests, policy learning, sensitivity analysis) and the inference rules shared across designs. Produces the design recommendation with the assumption that licenses it, the estimand and its subpopulation, R estimation code for the observables branch, and a drafted taxonomy paragraph. TRIGGER on "causal inference", "identification strategy", "which method", "research design", "endogeneity", "quasi-experiment", "natural experiment", "observational study", "selection on observables", "unconfoundedness", "conditional ignorability", "matching", "propensity score", "doubly robust", "AIPW", "double machine learning", "DML", "causal forest", "CATE", "policy learning", "targeting", "sensitivity analysis", "omitted variable bias", "Oster bounds", "coefficient stability", "overlap", "trimming", "marketplace experiment", "surrogate index", "mediation", "mediation analysis", "indirect effect", or any "how do I estimate the effect of X on Y" question with no design chosen yet. Once a design is chosen, the method skills own it: field-experiment, did, synthetic-control, rdd, iv.

Build and maintain the semester course website that a set of Quarto lecture decks hangs off: the landing page students bookmark, the week table, the site theme, the navbar and school mark, the instructor line, the handout post-render hook, and publishing to GitHub Pages. TRIGGER on "course website", "course site", "syllabus page", "course landing page", "week table", "add week 4 to the site", "link the new deck on the site", "publish the course site", "GitHub Pages for my course", "the site navbar", "logo on my course page", "put my syllabus online", "the site looks wrong on a phone". Use teaching-lecture to write or repair a deck; this skill owns everything around the decks.

Design, analyze, and write up randomized experiments (field experiments, A/B tests, RCTs), covering stratified and clustered designs, randomization inference, covariate adjustment done right, noncompliance, attrition and gated outcomes, treatment-effect heterogeneity, and interference. Produces advice with citations, R estimation and diagnostics code, and a drafted methods paragraph. TRIGGER on "A/B test", "randomized experiment", "RCT", "field experiment", "holdout", "lift test", "randomization inference", "stratified randomization", "cluster randomized", "geo experiment", "power analysis", "MDE", "encouragement design", "noncompliance", "ITT", "attrition", "Lee bounds", "CUPED", "variance reduction", "uplift", "heterogeneous treatment effects", "causal forest" (experimental heterogeneity; observational causal forests belong to causal-design), "interference", "spillover", "SUTVA". Pre-registration documents belong to the preregister skill; design triage across methods belongs to causal-design.

Author a Quarto reveal.js deck for an academic research presentation: conference talk, seminar, job talk, brown bag, workshop. Restrained styling, assertion titles, one argument per deck, exhibits from R or Python, deep appendix for questions. TRIGGER on "make slides for my talk", "build a seminar deck", "conference presentation", "job talk", "brown bag", "turn this paper into slides", "slides for my paper", "deck for the referee's question", "add an appendix slide", "export a handout". Use teaching-lecture for classroom material, and slide-review to audit a deck that already exists.

Render a Quarto reveal.js deck, screenshot every slide in a real browser, and review the pictures for overflow, unreadable type, low contrast, broken figures, failed math, and a weak argument, on a light talk deck or a dark lecture deck, and check that an offline-variant deck still works with no network. TRIGGER on "review my slides", "check my deck", "audit my slides", "pre-talk check", "pre-flight my deck", "am I ready to present", "are my slides readable", "will this deck work offline", "does my math render", "is anything cut off", "did my slides overflow", or before a seminar, job talk, conference talk, or lecture. Works on .qmd or an already-rendered .html; handles light and dark themes. macOS toolchain: quarto, deck-check.mjs on node 22, Playwright MCP, pdfread.py.

Design, estimate, validate, and write up a synthetic control analysis, including synthetic difference-in-differences, augmented/penalized variants, and factor-model/matrix-completion relatives for panel counterfactuals. Produces advice with citations, R estimation and diagnostics code, and a drafted methods paragraph. TRIGGER on "synthetic control", "synthetic DiD", "SDID", "donor pool", "comparative case study", "weighted counterfactual", "geo holdout", "CausalImpact", "matrix completion", "interactive fixed effects", "gsynth", "generalized synthetic control", or any setting where one or a few aggregate units (a state, market, DMA, category, platform region) got treated and untreated units must form the counterfactual. Staggered-adoption panels with many treated units belong to did; design triage across methods belongs to causal-design. Prospective geo experiments (choosing treatment markets by design) belong to field-experiment.

Author Quarto reveal.js lecture decks for a classroom, built for engagement: incremental reveals, animation, discussion prompts, checks for understanding, worked examples. Teaching blocks, staged reveals, and figures sized for a projector. TRIGGER on "build a lecture", "slides for my class", "teaching deck", "lecture on X", "course slides", "class on AI", "make a lecture for week 3", "MBA lecture", "undergrad lecture", "add a discussion slide". Use research-talk for a conference, seminar, or job talk, slide-review to audit a deck that already exists, and course-site for the landing page, the week table, and publishing.

Compile a TikZ diagram, rasterize it to PNG with ghostscript, have the tikz-reviewer subagent actually look at the image, apply the surgical fixes it returns, and repeat until APPROVED or 5 rounds elapse. TRIGGER when asked to polish, fix, debug, or iterate on a TikZ figure, a Beamer diagram, a .tikz or .tex snippet, or a DiD / parallel-trends / event-study / DAG / coefficient-plot figure whose labels overlap or that "looks wrong", and on "compile and review this tikz". macOS toolchain: latexmk from MacTeX plus gs.