Skip to content
Document conversion guide · facts checked 13 November 2026

C-CDA to FHIR conversion, without ten copies of every allergy

How to turn CCDs, discharge summaries and referral notes into FHIR R4 resources you can trust: the HL7 C-CDA on FHIR mappings, which sections each converter actually handles, what a conversion drops, how to merge repeated documents, and how to prove the output is right.

Surgical team in masks and caps looking down at the camera from around an operating theatre light
At a glance
  • C-CDA on FHIR 2.0.0HL7 STU 2, June 2026: twelve C-CDA to FHIR maps
  • Coverage checked per sectionSeven converters compared on what they really map
  • De-duplication by designRepeated documents merge instead of piling up
  • Validated both endsC-CDA in, US Core profiles out, counts reconciled

How do you convert C-CDA to FHIR?

Parse the document header into Patient, Encounter and author resources, then convert each section’s coded entries to FHIR R4 resources using HL7’s C-CDA on FHIR 2.0.0 mappings: allergies to AllergyIntolerance, problems to Condition, medications to MedicationRequest, results to DiagnosticReport and Observation, vital signs to Observation. Keep the original XML as a DocumentReference, translate OIDs to FHIR system URIs, match entries already loaded from earlier documents before creating anything, and validate against the US Core version your receiver uses.

Documents are not messages

Every CCD tells you everything again

A C-CDA document is a snapshot. A primary care practice that sends a Continuity of Care Document after every visit sends the full problem list, medication list and allergy list each time, plus a narrative a clinician can read. Load fifty of those naively and a patient has fifty penicillin allergies, some active and some resolved, with no way to tell which statement is current.

This page covers the document side of FHIR conversion as of 13 November 2026: what HL7’s C-CDA on FHIR guide maps and what it deliberately leaves out, a section-by-section table with LOINC section codes and target resources, seven converters compared on the sections they actually handle, what a conversion loses, de-duplication across documents, identifiers and code systems, a worked allergy example, validation, and the effort a real project takes.

For HL7 v2 feeds (ADT, ORU, ORM) the problems are different, and our HL7 to FHIR conversion guide covers them segment by segment. C-CDA work sits inside our FHIR interoperability and HIE integration services.

What is inside a C-CDA document

A C-CDA document is one XML ClinicalDocument with a header and a body. The header says who the patient is (recordTarget), who wrote it (author), which organisation holds it (custodian), and which encounter or period it covers (componentOf or documentationOf). The body is a list of sections. Each section has a LOINC code, one or more templateId values naming the rules it follows, a human-readable text block, and zero or more coded entry elements.

The document type comes from the header’s templateId and LOINC code: CCD, Discharge Summary, Referral Note, Progress Note, History and Physical, Consultation Note and others. They share section templates, so a converter that handles CCD sections handles most of a discharge summary too, except the sections only that document type uses, such as Hospital Course or Discharge Diagnosis.

Versions matter. Certified US EHRs build documents to C-CDA R2.1 with ONC’s Companion Guide Release 4.1, required under the HTI-1 rule from the start of 2026 and aligned to USCDI v3. HL7 has since published newer C-CDA editions, most recently 5.0.0 in June 2026, but most documents you receive today follow R2.1. Inside one document the same section can carry two template versions (an original and a dated extension such as 2015-08-01), and older senders still emit 2014-era templates. A converter that matches on one exact templateId silently skips the rest.

What HL7’s C-CDA on FHIR guide gives you, and what it leaves out

The C-CDA on FHIR Implementation Guide reached version 2.0.0 (STU 2) on 5 June 2026. Three things about this release are worth knowing before you plan:

  • It is now a mapping guide. Release 2.0.0 removed the FHIR document profiles that earlier versions carried. What remains are C-CDA to FHIR maps for twelve areas (allergies, encounters, immunizations, medications, notes, participation, patient, problems, procedures, results, social history, vital signs), FHIR to C-CDA maps for seven, structural guidance for data types, and terminology ConceptMaps.
  • The C-CDA to FHIR maps carry multi-vendor consensus. They were agreed through Connectathons with Oracle (Cerner), Diameter Health (now Availity), Google, MDIX, Redox, Smile Digital Health and the VA. The reverse direction has been piloted by one company only.
  • Missing maps are deliberate. If a C-CDA element has no row, the authors could not agree a target. Plan of Treatment, Goals, Health Concerns, Payers, Family History, Functional and Mental Status, Medical Equipment and Advance Directives have no mapping page at all. You have to decide those yourself, and write the decision down.

The guide is scoped to C-CDA R2.1 and US Core, which means its targets are US Core profiles. If your receiver uses International Patient Summary or India’s ABDM profiles, the clinical logic still helps but the profiles and terminology bindings differ.

C-CDA sections to FHIR resources

Section codes are LOINC. “HL7 map” means the C-CDA on FHIR 2.0.0 guide has a mapping page for that area; “your decision” means it does not.

Section (LOINC)Main entry templateFHIR R4 targetHL7 map?Watch for
Allergies 48765-2Allergy Concern Act wrapping an Allergy Intolerance ObservationAllergyIntoleranceYesThe concern act status and the observation status disagree in many exports. Negated “no known allergy” entries need the guide’s SNOMED CT map.
Problems 11450-4Problem Concern Act, Problem ObservationCondition (problem-list-item)YesOnset comes from the observation’s effectiveTime/low; the concern act’s dates describe when it was tracked, not when it started.
Medications 10160-0Medication Activity (substanceAdministration)MedicationRequest, per US CoreYesThe guide chose MedicationRequest even for a home medication list. Sig text, dose ranges and frequency live in nested structures.
Immunizations 11369-6Immunization ActivityImmunizationYesnegationInd plus a refusal reason becomes status not-done with statusReason. CVX codes, lot numbers.
Procedures 47519-4Procedure Activity Procedure, Act or ObservationProcedureYesThree entry shapes for one concept; moodCode INT means planned, not done.
Results 30954-2Result Organizer, Result ObservationDiagnosticReport plus ObservationYesThe organizer becomes the report and each component a result. Reference ranges and interpretation codes per result.
Vital signs 8716-3Vital Signs Organizer, Vital Sign ObservationObservation (vital-signs profiles)YesSystolic and diastolic arrive as two observations but US Core wants one blood pressure Observation with two components.
Social history 29762-2Smoking Status, Social History Observation, othersObservationYesSmoking status has its own US Core profile and value set; birth sex is a Patient extension, not an Observation.
Encounters 46240-8Encounter ActivityEncounterYesThe header’s encompassingEncounter and the section’s entries can describe the same visit twice.
Notes (several note-type codes)Note ActivityDocumentReferenceYesNarrative-only sections such as Assessment or History of Present Illness are often converted to notes too.
Plan of Treatment 18776-5Planned Act, Encounter, Procedure, Medication, ObservationServiceRequest, CarePlan, MedicationRequest (intent plan)Your decisionPlanned items share templates with done items; moodCode separates them.
Goals 61146-7, Health Concerns 75310-3Goal Observation, Health Concern ActGoal; Condition (health-concern)Your decisionOften narrative only. Goals reference concerns through entryRelationship.
Payers 48768-6Coverage Activity, Policy ActivityCoverage, OrganizationYour decisionMember IDs and subscriber relationships need care; many receivers do not want payer data from documents at all.
Family history 10157-6Family History OrganizerFamilyMemberHistoryYour decisionOne organizer per relative; conditions sit inside it.
Functional 47420-5 and mental status 10190-7Assessment scale and status observationsObservationYour decisionScores without the scale’s LOINC panel lose their meaning.
Medical equipment 46264-8, advance directives 42348-3Non-Medicinal Supply, Advance Directive ObservationDevice or DeviceUseStatement; DocumentReference or ConsentYour decisionImplantable devices carry UDIs; advance directives usually point to an external document.

The header maps too: recordTarget to Patient (race, ethnicity and birth sex as US Core extensions), authors and performers to Practitioner, PractitionerRole and Organization, and the guide recommends a Provenance resource for each converted entry, populated from its author, whether stated on the entry or inherited from the section or document (CDA calls this context conduction).

CCD to FHIR converters compared by section coverage

Checked against each project’s source code, package registry or documentation on 13 November 2026. “Sections” means what the converter maps to discrete resources, not what it can store as a note.

ConverterLicence and statusSection coverageOutput and caveats
Microsoft FHIR-Converter (Liquid templates)MIT. Commits to November 2026; last tagged release v7.0, June 2024.Nine document templates (CCD, Discharge Summary, Referral Note, Transfer Summary, Progress Note, H&P, Consultation, Operative and Procedure Notes) and 30 section templates, including Advance Directives, Family History, Functional and Mental Status, Goals, Health Concerns, Medical Equipment, Nutrition and Payers.Broadest open-source coverage. Templates predate the 2.0.0 guide, so check medications, nullFlavor and time-zone handling against it.
Azure $convert-dataOperation on Azure Health Data Services; documentation dated August 2026.Same templates as above.Returns a batch Bundle. Microsoft says the default templates are not meant for production and may change; host your own copy.
Metriport fhir-converter (Handlebars)AGPL-3.0 for the open-source parts. Last public commit 28 August 2026.Descended from Microsoft’s earlier Handlebars converter; C-CDA R2.1 in, batch Bundle out; used in Metriport’s own network data pipeline.AGPL obligations apply if you modify and offer it as a network service. Check the licence with your counsel first.
Medplum @medplum/ccda (TypeScript)Apache-2.0. Version 5.2.17, released 11 November 2026.Allergies, problems, health concerns, medications, immunizations, procedures, social history, encounters, goals, plan of care and care team. Result and vital-sign organizers become Observations with members; it creates no DiagnosticReport. Its source marks Notes, Payers and Reason for Referral as not yet implemented.Throws on an unrecognised section unless you set ignoreUnsupportedSections; good for catching gaps early. Builds a Composition alongside the resources.
ccda-to-fhir (Python, Nurra Health)MIT. Version 0.3, July 2026; an early-stage project.Patient, problems, allergies, medications, immunizations, procedures, results, vitals, social history, encounters, care plans and goals, care teams, payers, plan of treatment, notes and devices; states it follows C-CDA on FHIR 2.0.0.Outputs FHIR R4B, not R4: confirm your server accepts it. Raises an error on unknown code systems or unmapped values instead of dropping them.
Aidbox C-CDA converter (Health Samurai)Commercial, part of the Aidbox FHIR server.About fifty section templates, including admission and discharge diagnoses, hospital course, operative note sections and DICOM object catalog. Its own list shows Assessment and Plan, Physical Exam and Reason for Referral without a section template.Extensible with your own section rules. Tied to the Aidbox platform.
AWS HealthLake Data Transformation AgentPreview feature of HealthLake.An “AWS Starter Profile” of common section-to-resource mappings that you refine with an AI agent; also converts CSV.Output lands in a HealthLake data store. Preview status and generated mappings both mean you test it like your own code.
CareEvolution Orchestrate ConvertCommercial API; free sandbox; price on request.CDA to FHIR R4 as a stateless API call, alongside HL7 v2 and X12 conversion.No published section list we could find; test with your own documents in the sandbox.

Two practical conclusions. First, every option needs your sample documents run through it before you choose: coverage on paper says nothing about the template versions and quirks your senders produce. Second, the converter is the smaller half of the project. Identity, de-duplication, terminology and validation are yours with any of them. We have no commercial relationship with any vendor in this table.

What a C-CDA to FHIR conversion loses

  • Narrative-only content. Sections with text and no entries (Assessment, Physical Exam, Review of Systems, many Plan of Treatment sections) produce no discrete resources. Keep them as DocumentReference notes or they vanish.
  • Disagreement between narrative and entries. The human-readable table can show a dose change that the coded entry does not. A converter reads the entries; the clinician read the table.
  • originalText pointers. Many codes point into the narrative with <reference value="#allergy1"/>. Unless you resolve the reference, CodeableConcept.text ends up empty and the sender’s own wording is lost.
  • Time-zone offsets. C-CDA timestamps may omit the offset; FHIR requires one whenever a time is present. The guide lists the options: drop the time, borrow the offset from elsewhere in the same entry, section or document, infer it from the sender’s location, or default to UTC. Pick one per sender and record it.
  • Null flavours. CDA distinguishes “no information” (NI) from “unknown” (UNK); the guide maps both to FHIR’s unknown. The distinction is gone after conversion.
  • Nested detail. Allergy reactions and severities, medication indications and patient instructions, and problem status observations sit in entryRelationship elements several levels deep. Templates that stop at the first level drop them without an error.
  • Translations. A problem coded in SNOMED CT with an ICD-10-CM translation should become two codings in one CodeableConcept. Some converters keep only the primary code.
  • Document context. Which encounter the document described and who signed it come from the header. Without Provenance, a converted medication cannot say which document asserted it.

De-duplication across documents

This is the part most converters leave to you, and where most projects go wrong. The goal is one AllergyIntolerance per real allergy, however many documents mention it.

  1. Use the sender’s entry IDs first. Each entry should carry a stable id (root plus extension). Map it to an Identifier and write with a conditional update on that identifier. When the same EHR sends the same allergy next month, it updates the existing resource.
  2. Test whether the IDs really are stable. Some systems generate fresh UUIDs on every export. Compare the same patient’s documents from two dates before trusting IDs; if they change, fall back to a clinical key.
  3. Clinical key as the fallback. Patient plus normalised code plus onset date for problems; patient plus substance code for allergies; patient plus RxNorm ingredient plus start date for medications. Normalise to the standard code first, so a local code and its SNOMED CT translation match.
  4. Different senders, different IDs. The hospital and the clinic both record the same diabetes with their own identifiers. Matching across senders needs the clinical key, and a clinician-approved rule for which sender wins on status and dates.
  5. Absence is not resolution. If a newer CCD omits a problem, that does not prove it resolved: the sender may have filtered by date or by encounter. Only change status on an explicit statement.
  6. Keep the trail. Each source document becomes a DocumentReference; each resource gets Provenance entries pointing to every document that asserted it. Reviewers can then see why the record says what it says.

For US certified EHRs, ONC’s clinical information reconciliation criterion already expects users to reconcile problems, medications and allergies from received documents. If your product feeds clinicians, a review queue for conflicting entries is often better than automatic merging.

Identifiers and code systems: OIDs to URIs

CDA identifies code systems with OIDs; FHIR wants URIs. The guide is explicit that a known system must be translated, never wrapped: LOINC is http://loinc.org, not urn:oid:2.16.840.1.113883.6.1. Only OIDs with no known URI get the urn:oid: prefix. A starter table:

Code systemOID in C-CDAFHIR system URI
LOINC2.16.840.1.113883.6.1http://loinc.org
SNOMED CT2.16.840.1.113883.6.96http://snomed.info/sct
RxNorm2.16.840.1.113883.6.88http://www.nlm.nih.gov/research/umls/rxnorm
CVX (vaccines)2.16.840.1.113883.12.292http://hl7.org/fhir/sid/cvx
ICD-10-CM2.16.840.1.113883.6.90http://hl7.org/fhir/sid/icd-10-cm
UCUM (units)2.16.840.1.113883.6.8http://unitsofmeasure.org
NPI (identifier)2.16.840.1.113883.4.6http://hl7.org/fhir/sid/us-npi

Identifiers follow similar rules. An id with root and extension becomes system plus value; an id with only a root (an OID or UUID that is unique by itself) becomes system urn:ietf:rfc:3986 with the root, prefixed urn:oid: or urn:uuid:, as the value. Keep a per-sender table of the OIDs they use for MRNs and local codes, because those never appear in public registries.

Worked example: one allergy entry

Illustrative and abridged; the identifiers are invented. Templates and codes follow C-CDA R2.1.

<entry typeCode="DRIV">
  <act classCode="ACT" moodCode="EVN">
    <templateId root="2.16.840.1.113883.10.20.22.4.30" extension="2015-08-01"/>
    <id root="36e3e930-7b14-11db-9fe1-0800200c9a66"/>
    <code code="CONC" codeSystem="2.16.840.1.113883.5.6"/>
    <statusCode code="active"/>
    <effectiveTime><low value="20240312"/></effectiveTime>
    <entryRelationship typeCode="SUBJ">
      <observation classCode="OBS" moodCode="EVN">
        <templateId root="2.16.840.1.113883.10.20.22.4.7" extension="2014-06-09"/>
        <id root="2.16.840.1.113883.19.5.99" extension="ALG-4471"/>
        <code code="ASSERTION" codeSystem="2.16.840.1.113883.5.4"/>
        <statusCode code="completed"/>
        <effectiveTime><low value="20240312"/></effectiveTime>
        <value xsi:type="CD" code="419511003" codeSystem="2.16.840.1.113883.6.96"
               displayName="Propensity to adverse reaction to drug"/>
        <participant typeCode="CSM"><participantRole classCode="MANU">
          <playingEntity classCode="MMAT">
            <code code="70618" codeSystem="2.16.840.1.113883.6.88" displayName="Penicillin">
              <originalText><reference value="#allergen1"/></originalText>
            </code>
          </playingEntity>
        </participantRole></participant>
        <entryRelationship typeCode="MFST" inversionInd="true">
          <observation classCode="OBS" moodCode="EVN">
            <templateId root="2.16.840.1.113883.10.20.22.4.9" extension="2014-06-09"/>
            <id root="2.16.840.1.113883.19.5.99" extension="RXN-4471"/>
            <code code="ASSERTION" codeSystem="2.16.840.1.113883.5.4"/>
            <statusCode code="completed"/>
            <value xsi:type="CD" code="247472004" codeSystem="2.16.840.1.113883.6.96"
                   displayName="Wheal"/>
          </observation>
        </entryRelationship>
      </observation>
    </entryRelationship>
  </act>
</entry>

The converter writes one AllergyIntolerance, using a conditional update on the observation’s identifier so the next document from this sender updates it rather than adding a copy:

{
  "fullUrl": "urn:uuid:5b1e9f4a-2c7d-4e0b-a6f3-8d2c1e7b9a04",
  "resource": {
    "resourceType": "AllergyIntolerance",
    "identifier": [{ "system": "urn:oid:2.16.840.1.113883.19.5.99", "value": "ALG-4471" }],
    "clinicalStatus": { "coding": [{
      "system": "http://terminology.hl7.org/CodeSystem/allergyintolerance-clinical",
      "code": "active" }] },
    "category": ["medication"],
    "code": {
      "coding": [{ "system": "http://www.nlm.nih.gov/research/umls/rxnorm",
                   "code": "70618", "display": "Penicillin" }],
      "text": "Penicillin"
    },
    "patient": { "reference": "urn:uuid:0f6a3c2e-9b8d-4a71-b5e2-3c4d7f1a8e65" },
    "onsetDateTime": "2024-03-12",
    "reaction": [{ "manifestation": [{ "coding": [{
      "system": "http://snomed.info/sct", "code": "247472004", "display": "Wheal" }] }] }]
  },
  "request": {
    "method": "PUT",
    "url": "AllergyIntolerance?identifier=urn:oid:2.16.840.1.113883.19.5.99|ALG-4471"
  }
}

The decisions behind it:

  • The observation’s id had a root and an extension, so the root became the system (with urn:oid:, as it is not a registered URI) and the extension the value. The concern act’s root-only UUID is kept on the Provenance instead.
  • RxNorm’s OID became its FHIR URI. The #allergen1 reference was resolved against the section narrative to fill code.text.
  • 419511003 (propensity to adverse reaction to drug) set the category to medication through the guide’s ConceptMap; it does not prove an immune mechanism, so type was left out rather than guessed.
  • “20240312” stayed a date. No midnight time, so no time zone was invented.
  • A Provenance entry, not shown, records the author, the time and the source DocumentReference.

Validating converted documents

  1. Validate the input. Run incoming documents through a C-CDA validator, such as the one on ONC’s SITE test platform, on a sample from each sender. Knowing a sender’s errors up front explains half of the output problems later.
  2. Validate the output against the profile your receiver enforces. Load the US Core package that matches the receiving server’s CapabilityStatement into the HL7 FHIR Validator. Errors fail the build.
  3. Reconcile counts per section. Entries in, resources out, entries skipped with a reason. A section with ten entries and zero resources is a template mismatch, not an empty list.
  4. Replay a corpus. Keep a few hundred de-identified documents covering every sender, document type and template version, and convert the whole set on every change.
  5. Re-convert the same patient twice. Load two documents for one patient a month apart and assert that the resource count only grows by what actually changed. This catches unstable IDs and broken conditional updates.
  6. Clinician review. A clinician compares the rendered document with the converted record for a sample of patients, especially medications and allergies.

How a C-CDA conversion project runs, with effort

Illustrative, not client work: CCDs and discharge summaries from three senders, converted into a FHIR server that enforces US Core. Effort is in engineer-days and excludes the clinician’s review time.

PhaseWhat happensTypical effort
Document inventorySamples per sender, document types and template versions catalogued, sections counted, input validation run2–4 engineer-days
Converter choice and section decisionsShortlisted converters run on the samples; decisions for sections HL7 does not map3–5 engineer-days
BuildConverter configured or extended, OID and code system tables, Provenance, DocumentReference for originals6–12 engineer-days
Identity and de-duplicationID stability tests per sender, clinical keys, merge and review-queue rules4–8 engineer-days
Validation and corpusValidator in CI, per-section counts, repeat-document tests, clinician sample review3–6 engineer-days

Total: roughly 18–35 engineer-days over four to seven weeks. What stretches it: senders whose entry IDs change on every export, many narrative-only sections that a client wants as discrete data, and a receiving server on a US Core version nobody tested against early.

Packages

C-CDA to FHIR packages

Typical effort in engineer-days. We confirm a fixed price per phase after converting a sample of your real documents.

Converter bake-off

The problem

You have to choose a converter, and vendor claims about coverage are hard to compare.

How we do it

We run two or three shortlisted converters on your de-identified documents, count entries in and resources out per section, validate the output against your target profiles, and report what each one drops. Typical effort: 4–7 engineer-days.

What you get

  • Per-section coverage table on your data
  • Validation results per converter
  • Recommendation with the gaps you must fill
  • Sample corpus you keep
How it runs

How a C-CDA engagement runs

For one receiving system. More senders add inventory and identity work, not new stages.

Week 1

Inventory

De-identified documents from each sender, the receiving server and its profiles, and the clinician who will approve merge rules.

Weeks 1–2

Converter and decisions

Converters run on your samples, the choice agreed, and a written decision for every section HL7 does not map.

Weeks 2–5

Build and de-duplication

Pipeline, code system tables, Provenance, ID stability tests and matching rules, validation in CI.

Weeks 5–7

Parallel load and handover

Live documents converted alongside the current flow, counts and samples reviewed, then cutover, runbook and handover.

Senior engineers firstYour first call is with someone who will do the work, not a sales script.
You own everythingCode, configuration, reports and documentation stay in your accounts and are handed over.
We run our own products15 live platforms we build and operate, so we feel production pain too.
Honest about fitIf a cheaper route or another provider suits you better, we will say so.
FAQ

C-CDA to FHIR conversion: common questions

Yes. HL7’s C-CDA on FHIR Implementation Guide, version 2.0.0 published in June 2026, maps C-CDA to FHIR for allergies, encounters, immunizations, medications, notes, participation, patient, problems, procedures, results, social history and vital signs. It has no maps for plan of treatment, goals, health concerns, payers, family history or several other sections, so those decisions are yours.

The C-CDA on FHIR guide maps the Medication Activity entry to MedicationRequest, following US Core, which expects medication lists to be queryable through MedicationRequest even when the source system never held the order. Some converters still produce MedicationStatement, so check which one your receiver expects.

Among open-source options checked on 13 November 2026, Microsoft’s FHIR-Converter has the broadest template set, with 30 section templates and nine document types. Medplum’s library covers the core clinical sections, turns results into Observations rather than DiagnosticReports, and marks notes, payers and reason for referral as not yet implemented. Coverage on paper matters less than results on your own documents, so test with real samples.

Write each entry with a conditional update keyed on the sender’s entry identifier, after checking that the sender keeps those identifiers stable between exports. Where it does not, or where two senders describe the same condition, match on a clinical key such as patient, standard code and onset date. Never mark something resolved just because a newer document left it out.

No. Version 2.0.0 removed the document profiles that earlier releases carried and is now a mapping guide. If you need FHIR documents, use a Composition-based Bundle built to the profile your receiver specifies, such as International Patient Summary, or keep the original C-CDA as a DocumentReference.

Typically narrative-only sections, narrative wording referenced from codes, the difference between no information and unknown null flavours, nested detail such as allergy reactions or medication instructions when templates stop early, code translations, and document context unless you add Provenance. Keep the original document and log which entries were skipped.

For CCDs and one or two other document types from a few senders, our typical range is 18 to 35 engineer-days over four to seven weeks. De-duplication rules and the clinician’s sign-off usually set the pace. We confirm a fixed price per phase after converting a sample of your documents.

Both can convert C-CDA to FHIR R4. HealthLake’s Data Transformation Agent is in preview and starts from an AWS Starter Profile of section mappings that you refine. Azure’s $convert-data uses Microsoft’s Liquid templates, and Microsoft says the default templates are not meant for production. In both cases you still own identity, de-duplication and validation.

Last updated