Protocol

Eight components.Both columns.

A target trial protocol has eight parts. The engine writes each one twice: once as the trial that would answer the question, and once as the way that part would be carried out in observational data. Putting them side by side is what makes a disagreement between the two visible instead of buried.

Example question

Among adults admitted with acute decompensated heart failure, does early intravenous diuresis compared with delayed diuresis reduce 28-day mortality?

 
Target trial specification
Emulation using observational data
A
Eligibility criteria
The same criteria, judged using only information recorded before the start date.
B
Treatment strategies
Each strategy written as an observable pattern in the record, with the window in which it is classified.
C
Assignment procedures
Randomization replaced by adjustment, with the assumption that makes the groups comparable stated rather than implied.
D
Follow-up
Starts at assignment in both. The rules for when it ends are fixed before any data is pulled.
E
Outcomes
Each outcome given a definition the record can support, a measurement window, and a stated reliability.
F
Causal contrasts
The comparison and effect measure carried across intact, with the quantity being estimated named explicitly.
G
Identifying assumptions
What has to be true for the estimate to mean what it claims, which variables carry that weight, and what the absence of randomization costs.
H
Data analysis plan
What the analysis must account for, how missing data is handled, and which sensitivity analyses follow.
Structure

Every variablegets a role.

A list of variables is not a design. Each one is classified into one of nine causal roles, and that classification decides whether it belongs in the adjustment set. A confounder has to be in. A mediator must not be, because adjusting for it removes part of the very effect being measured. A collider must not be, because conditioning on it manufactures an association that was never there. The reasoning is recorded next to each variable in plain clinical language.

BASELINE SEVERITY KIDNEY FUNCTION AGE FRAILTY TREATMENT OUTCOME DIURETIC RESPONSE ICU ESCALATION ADJUSTMENT SET MEDIATOR COLLIDER UNMEASURED
Adjust Do not adjust Unmeasured Illustrative

Frailty is the honest part of this picture. It causes both treatment and outcome, it is rarely recorded in structured form, and no amount of adjustment recovers it. The engine names variables like this rather than leaving them out of the diagram.

9
causal roles, each with its own inclusion rule
14
bias mechanisms checked on every design
10
candidate start dates, weighed not assumed
21
reporting items tracked to completion
Reporting

Twenty-one items.Six are ours.

Published emulations are judged against a 21-item reporting checklist. Rather than claim coverage, the engine marks each item by what would unblock it. Six come out of the design itself. Seven are things only the investigator can supply. The remaining eight cannot be written until the analysis has been run.

Checklist items by what unblocks each one
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
6
written by the engine from the design itself: the causal question, the trial specification, its emulation, data sources, the abstract, and limitations
7
supplied by the investigator: background, rationale, ethics approval, registration, materials sharing, funding, conflicts
8
unwritable until results exist: participant counts, baseline distributions, follow-up, missing data, outcomes, estimates, sensitivity analyses, interpretation
Engine Investigator Post-analysis
Next
→ Boundaries What it will not do Continue →
Access

Demo ourservices.

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.

Request A Demo Contact