Knowledge map · the hidden spine of a research life
The surprising thing isn't the range of what's here — it's that one instinct runs underneath all of it: distrust the reported certainty, describe what's compatible, and make it reproducible. The domains below look unrelated; the published record says they aren't. And it didn't stay on paper — the same instinct rebuilt how an entire company reports its numbers.
Tap a node to light up its patterns — or tap a pattern to trace it back to its sources.
The field picked up one idea from you above all others — compatibility & surprise — and it's the hub of this entire map. The citation record and the structure agree on the center.
Your most-cited paper argues to replace confidence and significance with compatibility and surprise (BMC, 2020) — and it's also your single most frequent fight on #StatsTwitter: a p-value measures divergence from the tested model, not evidence for a hypothesis. Paper, blog, and timeline all carry one message.“P-values are not evidence for models, but rather evidence/information against the model you are testing.”
A published statistical methodologist — BMC, JAMA Psychiatry, an R package on Zenodo — running the fuel and market-basket BI of a convenience-store chain. Inference theory and c-store analytics in one résumé is a near-empty Venn overlap.
You co-authored "A Call for the Adoption of More Transparent Research Practices" (2019),
then lived it: concurve for consonance curves, statworkflow, Quarto with a freeze,
sessionInfo(), saveRDS over .RData. You argued the standard publicly,
then held yourself to it on a personal blog.
Read-only service accounts, hard plan caps to kill overage, dirty-read avoidance, IP whitelisting in triplicate (SQL Server + Azure SQL + firewall). You build redundancy the way a good estimator does — two independent ways to be right.
You train MMA — wrestling and striking, the most positional, control-the-variables corners of the sport. And your published Bayesian work is on the trained body itself: resistance training, overfeeding, composition (JFMK 2021; J Clin Densitometry 2021). Practice and method aren't two lives — they study the same thing.
This one isn't theoretical. The numbers leadership reads — fuel margins, market-basket, store performance — look the way they do because you rebuilt the reporting layer: a modern SQL Server → Snowflake → Power BI stack, held to the same skepticism and reproducibility you bring to a journal. The method didn't stay in the literature — it changed how the business sees itself.
Before BI stacks and S-values, the way in was nutrition and health science — wanting to know what actually works for the human body. Chasing honest answers there is what forced the statistics, and then the data career: overfeeding and body-composition studies (JFMK 2021; J Clin Densitometry 2021) sit right at that origin. The methodologist grew out of a question about diet and training.
The whole map runs on doubt, but yours has a discipline most contrarians lack: it refuses the second failure mode — cult-like skepticism that dismisses a claim without ever weighing it. Agnostic across paradigms, you grade evidence rather than gatekeep it: when a randomized trial is impossible, mechanism, observational data, and self-experiment still count as evidence, not noise to wave away.“We should use the best available evidence to make decisions… mechanisms, self experiments and observational data are the only things to work off of.”
concurve — consonance / compatibility curves and surprisal values. R
package, Zenodo, 2019.In public you're the corrector: precise, historically exacting, blunt about errors but rarely hostile — agnostic across paradigms, yet a fierce defender of nuanced frequentist thinking. Your most-shared threads are the same fight every time — the ASA / “retire significance” misreadings 97♥, the history of the p-value 89♥, and the p-value-as-divergence thread 52♥.