IRT Commandment #5
Thou Shall Think Long and Hard About Reconciliation
Reconciliation is a heavy resource activity that, under most circumstances, should not be undertaken.
This commandment is part of a larger, consistent story — using the IRT for its primary purposes (Commandment #1), understanding the nature of the data (Commandment 3), and sending very little data to the EDC (Commandment #4). If you internalize those principles, this one becomes obvious.
If you don’t, reconciliation becomes inevitable.
A Mindset Problem
If you recognize that the IRT is a tool, not a data collection or data management system, you’re already in the right mindset to understand why reconciliation should be rare. Many sponsors do not do it at all.
If, however, you think of IRT as an eCRF or as a secondary data management system, you will inevitably paint yourself into the reconciliation corner.
Once that happens, every difference looks like an error.
And every error demands explanation.
What IRT Must Be the Source Of
There are typically only two data elements that the IRT must be considered the source of — because it is the place where the data is first created:
Randomization (date, time, and treatment assignment)
Assignment / allocation of IMP to the subject
These are system acts. They happen in IRT by definition.
Reconciling these data points against another system is unnecessary and, in many cases, illogical.
A Sensible Extension
Some sponsors choose to go further and designate IRT as the source of what was dispensed to the subject — not just what was assigned or allocated.
I support this approach.
It further minimizes reconciliation and clearly establishes the IRT dataset as the source of truth for IMP activities, rather than forcing EDC to play that role.
This is not about moving data into IRT for convenience. It’s about assigning ownership where it naturally belongs — which is central to deciding whether reconciliation is necessary at all.
Even Strata Doesn’t Always Need Reconciliation
I’ll go one step further.
You could reasonably argue that strata data doesn’t need reconciliation at all, because the data entered into IRT — correct or not — is the data that was actually used to perform the randomization. It determined in that moment the treatment assignment.
That makes it operationally true, even if it later turns out to be clinically incorrect.
The truth can live in EDC.
The action lives in IRT.
Those two things are not the same — and that distinction matters.
We’ll come back to this in Commandment #6.
The Takeaway
Reconciliation should be intentional, limited, and justified — not automatic.
If you design IRT for its primary purposes, understand the nature of the data, and limit integrations, most reconciliation simply disappears.
And that’s not a risk.
That’s maturity.
Next up: Commandment 6 - The Randomization Schedule Is Sacred & Shall Not Be Modified