Standardsthe base

The protocol stops being a document and becomes data.

That is not our idea. It is the Digital Data Flow blueprint the largest sponsors in the industry agreed on, and roughly nine in ten protocols still arrive as unstructured prose, which is the gap it exists to close. AccuraTrials is a full public supporter of it, and we build on its output rather than around it.

Stackfour layers

Execution rests on semantics, not the other way round.

Read it from the bottom. Nothing at a higher layer can assert anything the layer beneath it cannot account for.

L4ExecutionEDC, eConsent, ePRO, labs, safetyDeterministic rules generated from the layer below run the study and capture the data.
L3Study modelOne living, versioned model of the whole trialVisits, forms, endpoints, eligibility, dosing and windows, machine-readable and diffable across amendments.
L2Ontology and semanticsCDISC USDM · SDTM · CDASH · controlled terminologyEvery extracted concept is bound to a standard definition, so “SBP”, “systolic” and “blood pressure (systolic)” resolve to one thing.
L1Source documentsProtocol, amendments, study specifications, SOPsThe evidence base. Nothing enters the model without a pointer back to a specific line here.

Basewhat we build on

Five standards, and what each one is actually for.

FoundationOpenStudyBuilderOpen study definition repositoryOur base layer: an open clinical metadata repository holding study definitions and controlled terminology as a graph rather than a proprietary schema, which is what keeps a definition portable and inspectable.
ModelCDISC USDMUnified Study Definitions ModelThe structure our protocol model conforms to. Born in TransCelerate’s Digital Data Flow initiative and now stewarded by CDISC, it is a shared way to say what a study is so it can move between systems without being re-typed.
ProtocolICH M11Structured protocol templateThe harmonised electronic protocol template. As sponsors move to M11 structure, the document we read and the model we build converge, so extraction gets more certain rather than harder.
DataSDTM · CDASH · ADaMCDISC data standardsCollection and submission standards. Fields are mapped at build time, so an export comes out submission-shaped rather than needing a mapping project after database lock.
VocabularyControlled terminologyMedDRA · LOINC · SNOMED CT · UCUMThe vocabularies that make the model mean something. Events, lab tests, findings and units resolve to codes rather than free text, which is what makes automated checking possible at all.

The regulatory frameworks this runs inside

Authorsthe base layer

We did not write the foundation. Three of the largest sponsors did.

OpenStudyBuilder is theirs, in the open, already running trials — and in May 2026 the three of them jointly committed to building and maintaining it. None of that infrastructure runs a study. The blueprint describes the trial; something still has to execute it and prove it happened.

Novo Nordisk
$47.7B revenue, 2025Initiated OpenStudyBuilder and released it as open source. Runs it across its own trials.
Boehringer Ingelheim
€27.8B revenue, 2025The largest privately held pharmaceutical company in the world, and a joint contributor.
Bayer
€17.8B pharma revenue, 2025Joint contributor to the open study-definition platform.

HistoryDDF, 2020 on

How the industry got here.

  1. 2020Digital Data Flow launchesTransCelerate’s initiative, with one goal: move the industry from a protocol that is a document to a protocol that is data, flowing into every downstream system instead of being hand-transcribed into each one.
  2. 2022USDM and a reference implementationDDF publishes USDM and an open reference Study Definitions Repository, proving the concept end to end. In parallel, Novo Nordisk releases OpenStudyBuilder as open source.
  3. 2023CDISC takes stewardship, ICH M11 landsUSDM becomes a CDISC standard with versioned governance, and ICH M11 reaches the same problem from the regulatory side. The blueprint stops being a pilot.
  4. TodayAccuraTrials runs itWe are full public supporters of DDF and build on its output rather than around it. The blueprint describes the study; our job is the machine that executes it and proves it happened.

Nextthe ask

Look at a real model, on a real protocol.

Thirty minutes. Bring your own protocol and we will compile it while you watch.