Skip to main content
CDISC study build · Core

Protocol to Define-XML. SDTM mapping included.

Seven agents read your protocol and draft USDM, DMP, eCRF, SDTM, ADaM, and Define-XML — each with a Part 11 trail and a human sign-off. Study builds that used to take 68 days close in under two weeks.

How it works
app.trialnexus.ai · STUDY-A4471
Protocol PDFparsed
USDM v4.0emitted
DMP + eCRFdrafted
SDTM + ADaMmapped
Define-XMLvalidated
Pinnacle 21 validation0 errors
The pipeline

Seven agents. One audit trail.

Each agent is an isolated activity in Temporal — logged, versioned, and attributable. Nothing leaves the pipeline without a human approval unless you configure it otherwise.

01
Agent 01 of 07

Parser

The protocol PDF is the source of truth for every downstream artifact. Parser reads it, resolves the structure, and emits USDM v4.0 JSON. Every element maps back to the source section — so every downstream agent knows exactly where its instructions came from.

USDM v4.0PDF ingestSection provenance
USDM v4.0 Output Complete
ElementSectionStatus
Endpoints (4)§3.2Parsed
Arms (2)§2.1Parsed
Visit schedule (12)§6.1Parsed
Assessments (34)§7.2Review
I/E criteria (18)§4.1Parsed
02
Agent 02 of 07

DMP Writer

The DMP is the first artifact every sponsor DM team produces and the first place weeks get lost. DMP Writer fills every section it can answer from the protocol, surfacing only the decisions that genuinely need a human. Near-complete plan in hours, not weeks.

DMP generationTemplate-awareDecision surfacing
DMP Sections 5 of 7 ready
SectionPagesStatus
Objectives & Endpoints4pReady
Data Flow & EDC Config6pReady
Query Management3pReview
Database Lock Procedure5pReady
Roles & Responsibilities2pReady
Protocol Deviations3pReady
SAE Reconciliation4pReview
03
Agent 03 of 07

eCRF Builder

eCRF build is where study timelines slip: field-by-field decisions, branching logic, edit checks, library alignment. eCRF Builder derives CDASH-compliant forms directly from the visit and assessment structure Parser extracted. Every field traces to the protocol section that required it.

CDASH compliantODM-XML / JSONField provenance
eCRF Modules 4 of 6 complete
ModuleVisitsFieldsStatus
Demography112Done
Vitals824Done
AE / SAEAll31Review
Labs686Done
Concomitant MedsAll14Done
Primary Efficacy418Review
04
Agent 04 of 07

SDTM Mapper

A wrong domain assignment or missed controlled term is a Pinnacle 21 finding at submission. SDTM Mapper works against CDISC Library CT, confidence-scores every mapping, and routes only the ambiguous cases to your DM lead. Never the obvious ones.

CDISC CT compliantConfidence scoringHuman-routed exceptions
SDTM Mapping Queue 2 flagged for review
eCRF Field→ SDTMConfidence
AEDECODAE.AEDECOD
97%
VSPOSVS.VSPOS
88%
LBMETHODLB.LBMETHOD
81%
EGMETHODEG.EGMETHOD
71%
RSDTCRS.RSDTC
58%
05
Agent 05 of 07

ADaM Generator

ADaM is where the clinical story is built: the datasets your biostatisticians use to generate efficacy and safety outputs. ADaM Generator derives ADSL, ADAE, ADLB, and custom datasets from the SDTM layer, recording every derivation as a traceable rule.

ADSL · ADAE · ADLBDerivation rulesR / Python / SAS
ADaM Datasets 4 of 5 ready
DatasetRecordsDerivationsStatus
ADSL31247Ready
ADAE89123Ready
ADLB14,24031Ready
ADTTE31212Ready
ADRS1,24818Review
06
Agent 06 of 07

Query Candidate

Query aging is the most visible sign of a trial under pressure. Query Candidate scans live EDC data against your edit check library, proposes the query text, suggests the resolution, and routes it to the right coordinator — before a DM has opened a data review screen.

EDC integrationEdit check libraryPre-proposed queries
Query Candidates — 14 pending Live scan
SiteFieldIssueConf
003LBORRESMissing value
94%
007AESTDTCDate format
89%
012MHTERMVerbatim mapping
82%
003VSPOSOut of range
76%
07
Agent 07 of 07

Define-XML

Define-XML is the last artifact before submission and the most common source of a complete response letter delay. Define-XML Generator produces v2.1 with cSDRG narrative, validates against Pinnacle 21 rules inline, and surfaces findings before the package leaves the pipeline.

Define-XML v2.1Pinnacle 21 inlinecSDRG narrative
Pinnacle 21 Validation All checks passed
0
Errors
0
Warnings
24
Domains
847
Variables
1,204 codelists · All origins traceable to protocol
Results

Study builds that used to take 68 days.

0
Days to full pipeline from protocol PDF
0%
Decisions with audit trail and provenance
0
Pinnacle 21 errors on first submission
∞
Simultaneous studies one platform can run
Decision Queue

The work surfaces to a human, with context.

Every agent output carries its reasoning, its confidence, and the source passages it drew from. Approve, revise, or reject in one click. Your DM lead focuses on decisions that require expertise. Not translation work.

app.trialnexus.ai · Decision Queue · STUDY-A4471
Decision Queue
4 awaiting review
Live
SDTM94% conf
Map AEDECOD → MedDRA PT
Ambiguous: "adverse event" vs. "diagnosis"?
DMP Writer81% conf
Query resolution window: 3 business days
Protocol silent on timeline
eCRF Builder73% conf
VSPOS field marked required
Cannot derive from §7.2 assessment schedule
Define-XML79% conf
Time-to-event origin: use RFDTC per §9.1
What a pilot looks like

Four weeks. One protocol. End-to-end.

Week 1

Ingest + USDM

We load your protocol PDF and your current CRF build standards. Parser emits USDM. Your team reviews the structural conversion and confirms the endpoint and visit structure.

Week 2

DMP + eCRF

DMP Writer and eCRF Builder produce drafts. We configure the agency dial and edit-check library with your DM lead. Your team approves the forms before they go anywhere.

Week 3

SDTM + ADaM

Against a source dataset (yours or a synthetic one), SDTM Mapper and ADaM Generator produce submission-ready outputs. Confidence scores and provenance on every mapping.

Week 4

Review + handoff

Full walkthrough of the audit trail. Export Define-XML. Pinnacle 21 validation inline. You decide whether to continue to a paid engagement. Either way, you leave with your full artifact set.

Common questions

What DM leads ask about the CDISC pipeline.

Seven Core agents: Parser (Protocol → USDM v4.0), DMP Writer, eCRF Builder, SDTM Mapper, ADaM Generator, Query Candidate, and Define-XML Generator. Each run is an isolated activity with its own LLM call and its own audit trail. Outputs route through the Decision Queue for human approval.
Study builds that used to take about 68 days close in under two weeks. The compression is the seven agents drafting in parallel, not one person queuing artifacts. Define-XML is checked against Pinnacle 21 rules inline, before it leaves the pipeline.
No. Core is additive. We connect to Medidata Rave (ODM XML) and Veeva Vault CDMS today, and we can export ODM-XML for any system that imports it. Your EDC stays where it lives; Core produces the CDISC artifacts and the review trail around it.
We are architected for Part 11 and ICH E6(R3): immutable audit logs, e-signature on approvals, versioning on every artifact, role-based access. Validation packages ship with the pilot. See the Trust Center for the current state of third-party attestations — we do not claim certifications we do not hold.

Your protocol. Your pipeline. Your audit trail.

See Core build a DMP, eCRF, and SDTM map from your own protocol in a 30-minute working session. No slides. We start with your CRF.

Talk to us first