A Governed Superset of the Open Stack: CuRE at OHDSI 2026

We will be exhibiting at the OHDSI Global Symposium in New Brunswick this month, October 20-22. Before we get there, it is worth writing down the argument, because the argument is the part worth arguing with.

”We export to OMOP” is the wrong sentence

Walk an exhibit hall at any observational research event and you will hear the same claim from nearly every table: we export to OMOP. It is usually true. It is also usually a statement about the last step of a pipeline — a nightly job that lifts data out of a proprietary model and lands it, lossily and late, in a common one. The common model is where the data goes to be analyzed, not where it lives.

That arrangement has a predictable set of consequences. Provenance ends at the export boundary. A mapping decision made upstream is invisible downstream. When someone asks why a concept was assigned the way it was, the answer lives in a system that the person asking cannot open. And the round trip — analysis findings flowing back to influence capture — never really happens, because the export runs one way.

CuRE takes the other bet. OMOP is not the export target; it is the substrate. Data arrives as FHIR through Conduit, Conduct maps it to OMOP CDM v5.4 with the mapping decisions themselves recorded, and every product above — the eCRF, the analytics workbench, the patient portal, the safety surface — reads from that same governed record. There is no export step, because there is nothing to export from.

The part that is harder to fake

The stronger version of the claim is not about storage at all. It is about protocol.

CuRE Conduct exposes an OHDSI WebAPI-compatible source surface: source registration, vocabulary search, concept-set resolution, and CDM results summaries. Which means a stock ATLAS instance — yours, running where it runs today, configured the way you configured it — can register CuRE as a CDM source and work against it the way it works against any other source you maintain.

This matters for a reason that has nothing to do with us. The OHDSI community’s answer to the vendor problem has always been to own the tools: keep ATLAS, keep the vocabularies, keep the methods packages, and refuse to let anything become a black box. Any platform asking that community to adopt it has to answer the obvious question — what do I give up? Our answer is: nothing on the tools side. Keep your ATLAS. Keep your cohort definitions. What you gain is the governance layer that the open stack was never trying to provide: tenant isolation, a Part 11 audit trail, consent- and DUA-scoped execution, and hosting that someone else operates.

A governed superset, not a replacement.

What we will actually show

At the booth we will be running an ATLAS front-end against CuRE Conduct, on synthetic data, on the table: the source appearing in the list, a vocabulary search resolving, a concept set expanding, a cohort generating. It is a demo that either works in front of you or does not, which is the only kind worth doing.

Around it are four short walkthroughs, cut from one story on one synthetic patient — one each for an ATLAS shop, a registry operator, a methods researcher, and a pharma RWE or safety team. All of it is synthetic data. No real patient data, no production tenant, nothing near the booth that should not be.

You do not have to wait for October, and you do not have to talk to us first: the engines run as in-browser demos on our platform page right now, on synthetic data, with no login.

Come and argue with us. Table conversations at OHDSI are better than most conference talks, and we would rather leave New Brunswick with a correction than a compliment.

Find us: OHDSI Global Symposium, October 20-22 2026, Hyatt Regency New Brunswick. Or start at our OHDSI page.