Merging HL7 fields without counting pipes

I've been working on a broken ADT. I had a known-good message and a slightly wrong one, and I needed a corrected copy by taking a few fields from one into the other.

The usual way is miserable. You count pipes, you copy a value, you miss a caret, you invent a new bug. I have done that more times than I want to admit.

Firstly, what are you actually trying to do?

You are not merging documents. You are deciding, field by field, which version of a patient fact should win. Birth place should be Wellington, not London. Patient class should stay inpatient. A pending location that only exists on one side should be created rather than left off. The clinical meaning comes first. The pipes are just how it is written down.

I already showed how to see those differences in the browser. Tools, Compare, in HL7 Soup Online. Semantic list on the left, text on the right. This is the next step. Once you can see PID-23, how do you fix it without counting?

Each difference row has two small buttons. << and >>.

<< takes the right value into the left. >> takes the left value into the right. The arrow points at the side that will change. I keep forgetting that for the first ten seconds, then it does what it says on the tin.

Keyboard works too. Select a difference, left arrow or right arrow. You can walk the list quite quickly.

In my case the top bar said PID-23 Birth Place. Left: London. Right: Wellington. I wanted the left message to be the corrected one, so I hit << and London became Wellington. The list updated immediately.

If a path only exists on one side, << is disabled. You cannot take a right-hand value that is not there. The tooltip is honest about it: cannot take the right value because this path is not present on the right. >> then becomes something more useful. Take the left value into the right and create the missing HL7 structure.

That is the standout bit. It will pad out the fields or the segment so the path can exist. No hand-building a PV1 just so field 16 has somewhere to sit. I tripped up on this years ago in raw editors, adding pipes until the field "looked" like it landed in the right place. It often didn't.

That is also the part that feels different from the Windows HL7 Soup editor. Compare I can do in a few tools. This field-by-field merge, including creating the missing structure, is the browser one I keep coming back to.

Revert Block sits under the text compare. Undo the block you just merged if you picked the wrong way. I used it. More than once.

Keep going and the field rows roll up. Once enough of PID matches, the individual PID-5.2 and PID-23 rows collapse into a single PID CHANGED. A bit surprising the first time. The list you were working down changes as you go. Keep picking.

I started with about twenty-five semantic differences and drove them down to none.

The empty state says: No semantic differences were found between these two HL7 messages.

A few things this is not.

It is two-way only. There is no base message, no three-way compare, no conflict marker. You are the arbiter. Last write wins per field. If you needed a proper ancestor merge you are in the wrong tool, and I would rather it said so than pretend.

There is no "save merged as a new message" button. The merged text is just the left and right editors. Copy the result out yourself. Treat them as a working copy.

You still paste the right-hand side. You cannot pick two messages from the loaded list inside the dialog. Same limitation as the compare post.

And please do not paste PHI into the online app. Use synthetic data, or anonymize first. Real patient messages stay in desktop HL7 Soup, or in HL7 Soup Web if you want the browser editor with the data staying on your own machine. A share link puts the message in the URL after the #, so it is not sent to the server, but that is not encryption. I would still only share a fake.

If you want to click around on a synthetic ADT^A04 from the built-in samples, this opens one in HL7 Soup Online:

Open the sample ADT^A04 in HL7 Soup Online

One ADT is about 1.9 KB of URL. Fine for a sample. Painful if you tried to stuff a real batch in there. The message is in the link, so treat the link like the message.

Anyway. Known-good on one side, broken on the other, cherry-pick the fields that matter, copy the winner out. No pipe counting. That is the whole trick.

Comparing HL7 messages in a browser

I had two ADT^A04 messages this week. Same EMR, same hospital, same patient. One registered Sam Brown without a fuss. The other got rejected. If you put them side by side in Notepad they looked the same until you got to the street.

Firstly, what does that actually mean?

The one that worked had this PID:

PID|1||74232^^^MRN||Brown^Sam||19780429|M|||19 High \T\ Main St^^Albany^NY^12207||555-555-2004

The one that failed had this:

PID|1||74232^^^MRN||Brown^Sam||19780429|M|||19 High & Main St^^Albany^NY^12207

The simple answer is that one of them escaped the ampersand in "High & Main", and the other one typed it in raw. But that raises the next question: isn't that just a different way of writing the same street?

No. In HL7, & is the subcomponent separator. You can see it declared right there in MSH-2 as part of ^~\&. It is not a letter in the address. It is punctuation that splits a field into smaller pieces.

So when the sending system wrote High & Main and left the & alone, the parser stopped treating Main as part of the street. Main, or the rest of that street, now sits in a different location than it does in the working message. The working one wrote \T\, which is the official escape for an ampersand. Same letters on the screen. Different place in the message.

To me it feels like an awkward leftover from the original delimiter design. I can picture a perfectly ordinary street sign, and some interface from the early '90s treating the & as structure. However, it is what it is...

But then the important question is how to deal with it?

Well, now that I do this in HL7 Soup, things have become simpler (yeah, like everything else). Open the message you already have. Then Tools, Compare.

The left pane is the current message. Paste the other one on the right. That is the usual job. I put the working ADT on the left and the rejected one on the right.

The list on the left is SEMANTIC DIFFERENCES. Each row has a path, a tag, and a short description.

The street is the interesting one. The working message kept High \T\ Main St as a single street value. The rejected one split it. PID-11.1.1 Street or Mailing Address is ONLY ON RIGHT, value 19 High. PID-11.1.2 Street Name is ONLY ON RIGHT, value Main St. Those paths do not exist on the left, because \T\ never created the subcomponents. Same street on the screen. Different place in the message. That is the row that explains the rejection.

Click PID-11.1.1. The top bar says Street or Mailing Address. Left: (not present). Right: 19 High. You stop saying "something in the PID" and start saying Main is sitting in a different location.

The other differences are real too. The working message has a home phone in PID-13, 555-555-2004. The failing one stops the PID after the address, so that field is only on one side. I would mention it on the vendor call. PID-12 is empty on the left and not present on the right. That is a real difference as well. Some systems treat empty and missing as the same thing. Some do not. MSH-10 is just the two control IDs, OK001 versus BAD001. None of that is why the second one failed. The ampersand is.

The right side is EDITABLE TEXT COMPARE. Side by side, both sides editable, the usual red and green. Useful when you want to stare at the raw PID line. I still start with the semantic list. That is the bit that understands HL7 paths instead of characters.

Why start there?

Because a text compare will happily tell you that & and \T\ are different characters, then leave you to work out what that does to every field after it. I have counted those pipes in Mirth exports more times than I want to admit. One extra delimiter and you are staring at the wrong slot. Click PID-11.1.1 and it tells you the street moved. You do not have to remember which number the street is, and you do not have to re-count after someone stuffed an extra caret into the name.

If you want to try the same two messages, here they are. Working first, then the one that failed.

MSH|^~\&|EMR|CITYHOSP|ADT|CITYHOSP|202608191000||ADT^A04^ADT_A01|OK001|P|2.5
EVN|A04|202608191000
PID|1||74232^^^MRN||Brown^Sam||19780429|M|||19 High \T\ Main St^^Albany^NY^12207||555-555-2004
PV1|1|O

MSH|^~\&|EMR|CITYHOSP|ADT|CITYHOSP|202608191000||ADT^A04^ADT_A01|BAD001|P|2.5
EVN|A04|202608191000
PID|1||74232^^^MRN||Brown^Sam||19780429|M|||19 High & Main St^^Albany^NY^12207
PV1|1|O

Anyway. Two messages, one works, one doesn't, and Compare points at the unescaped street ampersand. That is the whole trick.

Next time I will cover the little << and >> buttons on each difference. Those let you cherry-pick a field from one message into the other without counting a single pipe.

 

Comparing HL7 Integration Engines: 2026 Review

Updated July 2026

Last year I compared some of the main HL7 interface engines and ended up choosing HL7 Soup Integration Host. A lot has changed since then. Mirth has completed its move away from open source, iNTERFACEWARE has taken a rather unusual new direction, and HL7 Soup Integration Host has a new name: Integration Soup.

My own thinking has changed a little as well. I used to give code-led engines extra credit simply because anything could be coded. After spending more time inside some of our older Lua interfaces, I am less convinced that writing everything ourselves is an advantage. Modern graphical engines are far more capable than they used to be, and they can remove a lot of repeated logging, connection, security and error-handling code without limiting the business logic.

Integration Engine	License & Cost	Ease of Use	Standards & Connectivity	Scalability & Deployment	2026 View
Integration Soup Previously HL7 Soup Integration Host	Commercial subscription with workflow, site and enterprise options. Pricing is by quote. A 30-day trial is available.	Very high. Visual workflows, drag-and-drop transformations and readable message views cover most ordinary work. C# remains available when code really is the clearest answer.	HL7 v2, FHIR, JSON, XML, CSV and now DICOM. MLLP/TCP, HTTP/S, REST/SOAP, OAuth, files and broad database support. Extension libraries add cloud, SFTP, validation, document and encryption activities.	On-premise or cloud deployment, with Windows and .NET Core builds. Azure and AWS extensions, Git-based workflow deployment and product, site and enterprise licensing options.	Strengths: The easiest product here to understand quickly. Version 4.0 added useful depth without losing the simple workflow model. HL7 Soup Web is becoming an excellent browser companion, and the migration tooling is a real advantage. Weaknesses: Smaller vendor and installed base than the enterprise incumbents. I would like to see more of the full Integration Soup experience move into the browser.
Infor Cloverleaf	Commercial enterprise product with quote-based pricing.	Medium. GUI tools and wizards are improving, but complex work can still require specialist Cloverleaf and TCL knowledge.	HL7 v2, FHIR, CDA, X12 and broad legacy connectivity. Includes API gateway, FHIR and Da Vinci capabilities.	Very high scalability and high availability. Available on-premise, in the cloud or as a hybrid deployment, including container options.	Strengths: Proven at large healthcare organisations, broad standards support and strong operational reliability. Weaknesses: Enterprise cost and complexity. It remains a product that benefits from experienced specialists rather than occasional interface developers.
iNTERFACEWARE Iguana / IguanaX	Commercial. Published pricing starts at US$10,000 per year for two integrations, with larger packages by quote.	Low to medium. IguanaX has a more visual component model, but Lua and code remain central to building and maintaining the interfaces.	HL7 v2, FHIR, JSON, XML, REST, databases, files and other formats through components, libraries and custom Lua.	Cross-platform deployment with cloud options. IguanaX development is active, with high availability and SSO identified as current areas of focus.	Strengths: Fast, flexible and still extremely capable when you have strong developers and want complete control. Weaknesses: My experience of the older codebase has exposed how much repeated plumbing the customer can end up owning. More importantly, the company now describes itself as being largely run by its founder. Whatever the theory behind that decision, it creates a support and business-continuity concern for a production interface engine.
Rhapsody Corepoint	Commercial enterprise pricing based on organisation size, deployment model and requirements.	Very high. Its graphical, analyst-friendly approach remains one of the strongest in the market, and Rhapsody Axon now adds integrated AI assistance.	HL7, FHIR, REST APIs and other healthcare formats, with reusable actions, testing, monitoring and alerting.	Enterprise scale with high-availability options. Available as SaaS, on customer infrastructure or as an iPaaS deployment.	Strengths: Excellent GUI, support and enterprise record. It was named Best in KLAS again in 2026 and is no longer fairly described as simply Windows-only. Weaknesses: Premium pricing and a closed platform. It can be more product than a smaller integration team needs.
Mirth Connect by NextGen Healthcare	Mirth Connect 4.6 and later are commercial and proprietary. Version 4.5.2 remains available as the final open-source release but is no longer maintained by NextGen.	Medium. The channel GUI is familiar to a very large community, but JavaScript or Java is still needed for a lot of real mapping and workflow logic.	HL7 v2, X12, JSON, XML, REST, SOAP, DICOM and commercial FHIR capabilities.	Cross-platform and cloud-agnostic, with commercial clustering, security extensions and central monitoring through Mirth Command Center.	Strengths: Huge installed base, a great deal of existing knowledge and very flexible scripting. The commercial product continues to be developed and supported. Weaknesses: The maintained free edition is gone. The dated development experience and amount of scripting are harder to excuse once commercial licensing is part of the decision.
Qvera Interface Engine (QIE)	Commercial with flexible subscription licensing and pricing based on requirements.	High. A browser-based visual channel editor handles the main workflow, with JavaScript available for more complex requirements. Qvera now also promotes an AI Companion for help, mapping and troubleshooting.	HL7 v2, FHIR, CDA, X12, DICOM, ASTM, databases, JSON, XML and web services.	Cloud-optimised and container-friendly, with AWS, Azure and Google Cloud deployment and orchestration options.	Strengths: Modern web interface, strong standards coverage, credible cloud deployment and a good reputation for support. It remains one of the strongest alternatives in this list. Weaknesses: Smaller market share than the major enterprise platforms, quote-based pricing and some scripting for complex work.

So... what do I choose in 2026?

The answer is still Integration Soup, although I think I have better reasons for choosing it this year.

In the past I tended to see a code-first engine as the more powerful option. If everything is code, then surely nothing is out of reach. The problem is that you also take responsibility for everything you code. Looking through our existing Iguana libraries, a surprising amount of the Lua is there to manage connections, logging, errors, security and cleanup. It is important work, but it is also repeated work, and every copy becomes another place for a mistake or an old security decision to survive.

Modern graphical engines do not have the same limitations that put people off them years ago. A good workflow designer can handle the normal integration work while still allowing code where it is genuinely useful. That leaves the developer concentrating on mappings, transformations and the rules that are actually different for that interface.

Integration Soup 4.0 has moved a long way in that direction. It now has broader database support, DICOM, Azure and AWS extensions, stronger source-control deployment and tools for converting existing channels. The rename from HL7 Soup Integration Host also makes sense: this is now much more than an HL7 utility with an engine attached.

I particularly like what is happening in the browser. HL7 Soup Web is already a very good way to inspect, understand and share a message, and Integration Soup is beginning to take over from there when that message needs to become a real workflow. I would definitely like to see more of the Integration Soup experience move into the browser, but the direction looks right.

Corepoint and Qvera remain very credible alternatives, and Cloverleaf is still difficult to argue with at true enterprise scale. For the sort of work I actually want to build and maintain, though, Integration Soup gives me the best balance: a proper interface engine, a modern GUI, code when I need it, and far less code that I have to own simply to keep an interface safe and running.

I’ve just found a very handy feature in HL7 Soup Online for sharing an HL7 message with someone else.

Right-click on any message in the list and select Copy message as link. This puts a hyperlink onto your clipboard which you can then paste into an email, support ticket or chat message.

Right-click on your message and select Copy message as Link

When the other person clicks the link, the exact message opens for them in HL7 Soup Online. They can see the readable description of the message, click through the fields, and find out what the values mean without having to install anything.

Take a look at this link to see it in action.

This is particularly useful when the person you are sending it to doesn’t work with HL7 every day. Pasting the raw message into an email normally gives them a block of text full of pipes and carets. Sending the link lets them see that it is a patient registration, which details are for the patient, what sort of visit it is, and where each value sits in the message.

I can see this being useful when asking a software vendor about a particular field, adding an example to a support ticket, or explaining a message to someone who understands the workflow but not necessarily HL7.

There is one important thing to remember. The message is included in the link, so you should treat that link in exactly the same way as you would treat the original message. Don’t copy a live message containing patient information and send it as it is.

HL7 Soup Online has an Anonymize menu at the top of the screen. If you are starting with a real message, anonymize it before creating the link. I would still have a quick look through the result afterwards to make sure that any local identifiers or unusual fields have also been dealt with. If possible, a synthetic test message is always the safest one to share.

That’s really all there is to it. Right-click the message, copy the link and send it. The recipient gets the actual message in a tool that helps them understand it, rather than having to work it out from the raw HL7 themselves.

Moving an Iguana Database Interface

Take a fairly normal Iguana interface that receives a message and writes some of it to a database.

The useful part is quite small. Read a few values from the message, convert them into the format the database expects, and insert or update the record.

But that isn’t all the Lua we ended up with.

We also need to open the database connection, make sure it is still available, handle the error if it isn’t, log enough information to work out what happened, and make sure everything is closed properly afterwards. If the interface needs to retry, that needs to be dealt with as well.

None of this is unusual code. In fact, that is really the point. It is code for exactly the same problems that every database interface has.

I have been looking at this while starting to move our Iguana channels to Integration Soup. When I create the database step there, the connection and the normal failure handling belong to the engine. I configure what I want to do and concentrate on the values I want to write.

The result is that most of the old Lua doesn’t need to be converted at all. It wasn’t business logic. It was just everything we had written around the business logic to make it safe to run.

I must admit, the large Lua libraries had always made the interfaces feel quite powerful. If we needed something, we could code it. That is still useful, but it also meant we took responsibility for a lot of things that would probably have been better handled by the integration engine.

It also makes later changes harder than they need to be. We have some old TCP interfaces where TLS now needs to be added. With our own implementations, each one has to be checked and updated. In Integration Soup it is a connection setting, so the same tested implementation is used throughout.

There will still be places where I use code. Some mappings and unusual rules are simply easier that way. But I would much rather use code for the bit that makes an interface different, and let the engine look after the bits that are the same every time.

That is probably the biggest improvement I have seen from the conversion so far. I’m not losing the useful code. I’m finding out how much of the other code I no longer need.

How Much Iguana Code Did We Actually Need

I’ve now started looking through some Iguana channels to see what will be involved in moving them to Integration Soup.

One of the first things I noticed was just how much Lua we have. I already knew some of the libraries were quite large, and I suppose I had always taken some comfort from that. All the code was there, we could see what it did, and if we needed the interface to do something different then we could change it.

What I hadn’t really considered was how much of that code had nothing to do with the actual interface.

There is code for logging, opening and closing database connections, handling errors and making sure that a failure doesn’t leave anything in a bad state. We have also written code around TCP connections, certificates and all the other things needed to keep an interface running properly.

Obviously these are all important. The problem is that they are important for every interface, and we have ended up writing much the same code over and over again.

If it is written again for every channel, there is always the chance that one version is going to be slightly different or have a mistake in it. If it is copied from another channel, that helps at the time, but it also means the same code is now scattered everywhere. When something needs to be improved later, every copy has to be found and changed.

You can move it into a common Lua library, and we have done that where it made sense, but then you have a large library that also needs to be maintained and tested. It is better than having lots of copies, but it is still our code and our problem.

This has been particularly noticeable with security. Some of our older TCP interfaces were created without TLS. That is not a situation they should still be in, but adding it means working through the old implementation and making sure the change is made correctly everywhere it is needed.

In Integration Soup the secure TCP/MLLP connection is already part of the product. I still need to configure the certificate and make sure the other system is set up correctly, but I am not having to write or maintain the code that provides the secure connection.

The same thing is happening with the rest of the conversion. Once Integration Soup is handling the connection, logging and normal error behaviour, a surprising amount of Lua is simply no longer required.

What I am left with is mostly the part I am actually interested in: which fields need to be mapped, what needs to be converted, and any business rules that are particular to that interface.

I can still use code when code is the best way to do something. The difference is that I am using it for the unusual part of the interface, rather than using it to provide all the normal features that I would expect the integration engine to handle.

I don’t think there was anything wrong with having Lua available in Iguana. It certainly meant we could get things working. I just don’t think the size of the codebase gave us quite the advantage I had assumed it did.

So far the converted work is smaller and much easier to follow. More importantly, there is less of our own code sitting between a message arriving and it getting where it needs to go.

Iguana to Integration Soup migration

 Migration brochures have a soothing habit of turning a working interface estate into three boxes and an arrow.

Export. Convert. Go live.

Anyone who has supported real HL7 channels knows there is usually rather more hiding inside the arrow. There are Lua modules shared by six channels, a VMD last edited by somebody who has left, environment variables with misleading names, certificates, special ACK rules and sample messages that explain more than the documentation.

The good news is that an Iguana migration does not have to begin from a blank page. Iguana's code-led structure gives us useful source material, and Integration Soup now has a practical process for turning that material into testable workflows.

What do you actually export from Iguana?

The short answer is: the channel exports are the start, not the whole package.

I would collect:

• Git channel exports or Translator project archives. These provide the channel structure and project files.
• All Lua code. Include local modules, shared modules, and any common libraries used across channels.
• VMD files. These often contain message definitions and mapping knowledge that is not obvious from the Lua alone.
• Channel and instance configuration. Source, filter and destination settings, environment variables, queue behavior, web settings and notification rules all matter.
• External dependencies. Record databases, APIs, file locations, certificates, custom executables, plugins and scheduled jobs.
• The channel topology. Document From Channel and To Channel relationships, fan-out, shared queues and any processing split across several channels.
• Representative test data. Include de-identified successful messages, known failures, unusual variants, ACKs and boundary cases. Do not put patient information into a migration repository.

If that sounds like more than a single export button, it is. The point is not merely to preserve the Lua. The point is to preserve the behavior.

How the Iguana pieces map into Integration Soup

An Iguana channel normally has a recognizable flow: receive something, filter or transform it, and deliver it somewhere. Integration Soup expresses the same intent as a workflow.

• The Iguana source becomes an Integration Soup receiver: LLP, HTTPS, file, database or the appropriate input.
• Filters and Translator logic become visual transformers, filters and code activities in the same processing order.
• The destination becomes a sender or workflow activity with its endpoint, payload, retries and response behavior.
• Lua functions and modules are translated into C# and Integration Soup APIs, keeping function boundaries and names recognizable where that helps maintenance.
• VMD mappings become message templates, mappings or code transformers, whichever leaves the intent clearest.
• Environment values become deployment configuration rather than values buried inside a workflow.
• Channel-to-channel routing becomes explicit workflow output and connected activity flow.

I like this approach because it does not pretend the two engines are identical. Iguana runtime calls have to change, and some designs should be simplified rather than copied line for line. The aim is to retain the business rules while producing a workflow that looks native in its new home.

The migration is proven offline

This is the important part.

A converted channel should not be accepted because the diagram looks right or because one cheerful ADT message reached the destination. It should be tested as a regression exercise against what the Iguana channel already does.

Start with the exported samples and de-identified log cases. Feed the same inputs through the reference behavior and the new Integration Soup workflow, then compare:

• the parsed source values and validation results;
• the generated outbound payloads and encodings;
• which routes and destinations were selected;
• ACK content, timing and negative-ACK behavior;
• database writes and queries;
• file names, API requests and other external effects;
• ignored-message and error paths;
• retry, alert, queue and recovery behavior;
• throughput and resource use at representative volumes.

External systems can be stubbed or redirected in the offline environment, so the workflow can be exercised without sending test patients into a production EHR or emailing a thousand imaginary lab results to somebody.

Sometimes the output can be compared byte for byte. Sometimes fields such as timestamps, message control IDs or generated file names are expected to differ. In those cases the comparison should normalize only the agreed dynamic values and remain strict about everything else.

That is what “100% accurate” should mean in a migration: every agreed behavior and test case passes. It should not mean that someone looked at two messages and felt optimistic.

Prove one difficult channel first

I would begin with a representative channel that contains real transformation logic and at least one dependency. This is much more valuable than converting ten relays to make the progress chart look impressive.

The pilot tells you how your organization uses Lua, shared modules, VMDs, environment settings and error handling. Once those patterns are understood, the rest of the estate becomes easier to estimate and convert consistently.

It also gives the support team a chance to use the Integration Soup dashboard, inspect messages, search logs, follow the workflow and decide whether the new system makes operational sense. A migration succeeds when the people on call can support it at 2 a.m., not when the project team completes a spreadsheet.

Cut over in controlled groups

After offline acceptance, I would move a small group of interfaces at a time:

1. Freeze or carefully track changes to the Iguana channel during final testing.
2. Apply production endpoints, credentials, and certificates through controlled environment configuration.
3. Run in parallel or shadow mode where the surrounding systems allow it.
4. Agree the cutover step, monitoring window and rollback trigger before touching production.
5. Switch the listener, route, port or upstream destination during the planned window.
6. Compare live inputs, outputs, ACKs, queues and alerts closely.
7. Keep the old path available for the agreed rollback period, without letting both systems create duplicate downstream effects.

For many channels, the final switch is genuinely simple. The work that makes it simple is the inventory, conversion, and offline proof completed beforehand.

Why I think this is a sensible direction

Integration Soup gives me the things I want from a serious interface engine: visible workflows, queues, logging, monitoring, alerts, scheduling, environment configuration, code when needed, and support for the mix of HL7, FHIR, files, databases, APIs, cloud services and DICOM that turns up in real integration work.

It is also simple enough that the migration does not finish with all the old Lua hidden inside a generic coding window. The workflow remains understandable and operationally useful in the browser, while the translated C# is there for the logic that deserves code.

Most importantly, the Iguana migration service is built around evidence. Export the channels and their dependencies, rebuild the intent, replay the test pack, compare the behavior, then cut over in controlled groups. Approved migrations can also qualify for price-matched licensing and included conversion assistance, subject to the published terms.

There is no magic migration button, and I would be suspicious of one. There is something better: a repeatable path that lets you prove the new interface before the old one stops carrying the messages.