What it does.What it will not.
A design that looks correct is worse than one that visibly fails, because only the second one gets caught. Most of the engineering here exists to keep failures visible rather than tidy them away.
What it does
What it will not do
01
Resolves each concept to a field in your schema, with its timing and a stated reliability.
Substitute the nearest available field when the right one is missing. An unresolved reference is reported as unresolved.
02
Declares every node and edge in the causal graph, then derives the adjustment set from it.
Claim the graph is correct, or that the effect is identified. Both remain your judgment to make.
03
Names the quantity being estimated and states what the analysis has to account for.
Prescribe a statistical method, or run one. Choosing the estimator stays with the analyst.
04
Specifies what to measure, over what window, and against which definition.
Produce effect estimates, participant counts, or any figure it did not read from your data.
05
Checks each bias mechanism and names the diagnostic to run once the data is pulled.
Assert the design is unbiased. What cannot be resolved by design is written down as unresolved.
06
Raises malformed or missing output where it happens, with the record of where it came from.
Quietly repair its own output into something that reads as complete.