Each decision constrains the next, so the engine works in order rather than all at once. The start date governs eligibility. Eligibility governs the comparison. The comparison governs what the analysis has to account for. Step through the sequence below.
Follow-up has to begin somewhere, and the wrong choice bakes in bias that no later analysis can remove. Each candidate is assessed on whether eligibility is knowable at that moment, whether treatment can be assigned there, whether follow-up can begin there, and what it risks. Discharge, for instance, looks convenient and quietly excludes everyone who died first.
Ten candidate anchors, each with its own scoring profile. A default is a starting point, not a verdict: the engine states which one it chose for your question and why the others were rejected.
Cohorts are usually assembled from whichever filters a query happened to contain, and the attrition is reconstructed afterwards if anyone asks. Emulation forces each criterion to be a stated rule, applied once, at a defined moment, with what it removed recorded as it happens.
Figures shown are invented for illustration. The engine writes the criteria and the rules for applying them; the counts come from running those rules against your own data.
TriEmulate is enterprise infrastructure under active development and is not yet generally available, but it runs today and can be walked through end to end on a question of your own. A United States provisional patent application covering the system was filed in September 2026. If you work on causal inference in health data, bring a question and we will put it through.