See your whole trial. Prove every data point
Decentralized trials spread patient data across EDC, IRT, eCOA, labs, wearables and home visits. SourceParity brings it into one independent view, proves every mismatch, and keeps your data ready for FDA submission and inspection
LXP-301 · Phase III · 12 sites · 184 subjects
Likely cause: visit date entered for the screening call, not the clinic visit. Next step: ask the site to confirm the visit date
Decentralized trials multiplied the sources. Reconciliation is still done by hand
a week per study spent on manual reconciliation
Veeva Clinical Data Industry Research, 2025of teams reconcile outside their clinical systems
Veeva, 2025on average from last patient visit to database lock
Tufts CSDDOne engine checks every source. Three answers come out
SourceParityReads exports, read-only. Checks every pair against your protocol rules on every loadOnly the sites where data does not match
Patient by patient, every day
Every finding and decision, ready to export
Send monitors where the data breaks. Skip the rest
Each month SourceParity names the sites that need a visit and the ones where every source agrees, with the reason for each
- Sites ranked by open mismatches per subject
- Late source feeds flagged next to the site
- Visits saved, counted at your own cost per visit
Lock patients as their data reconciles. Not all at once at the end
See who is clean, who is waiting and exactly what is holding each subject back
- Every blocker named, with its owner
- Projected lock date at the current pace
- Locked records stay sealed; reopening keeps the original
184 subjects
When the inspector asks, the answer is one export
Every finding keeps both readings, the rule, the protocol section, who decided and when. Nothing is overwritten
- Discrepancy listing for the whole study
- Machine-readable audit trail
- Version, lock and change control record
Evidence pack · LXP-301
Start with a study you have already closed
Send us a recently locked study. We check it across every source and show what manual reconciliation missed. No risk to a running trial, and a clear answer in weeks
Closed study · Phase II · 88 subjects
Example data
We are onboarding three design partners for early 2027
One closed study checked end to end, a fixed scope agreed upfront and direct access to the team building the system. Your feedback shapes what we build next
Built for regulated data. From the first line of code
We read the exports your systems produce. We never write back to EDC, IRT or lab systems
Subject and site numbers only. Treatment assignment is never shown
Every load, rule version, finding and decision is recorded and exportable
The rule engine is built for validation under GAMP 5. AI only explains findings, is assessed for fitness for use and never makes a GxP decision
Rules prove. AI explains. A person decides
Questions sponsors ask us
Do you need access to our EDC?
No. We work from the exports your EDC, IRT, labs and vendors already produce. Access is read-only and nothing is ever written back to your systems
Where is our data stored, and who can see it?
In the US, on AWS. Only the people you grant access to can see a study, limited by study, site and data type, and every view is logged. We hold no direct patient identifiers, only subject and site numbers
How long does a retrospective audit take?
Usually a few weeks from receiving the exports, depending on the number of sources and the size of the study. You get a report of every mismatch with its evidence
How do you use AI?
The checks are deterministic rules, built to be validated like any GxP software, so every finding can be shown to an inspector. AI explains findings in plain language and suggests next steps. Its output is reviewed by a person, the model version is fixed, and it never changes data or makes decisions
Is the system validated?
We are building the validation package under GAMP 5 and 21 CFR Part 11 from day one, and we share it with our customers. The AI layer is not validated like deterministic software: it is assessed for fitness for its limited, advisory use, with human review of every output
What happens on the 30-minute call?
No slides. You tell us how you reconcile data today, and we show you what we would check first in your study
From first patient to FDA submission, know your data is right
30 minutes. Tell us how you reconcile today, and we'll show you what we would check first