Skip to content

Data contracts with registration vendors that survive a change of platform

Unified dataUpdated 2026-08-186 min read

In short

A data contract with a registration vendor is a written, testable agreement about the shape and timing of a feed. It names the primary key, the required columns and types, the arrival window and the notice period for changes, so the agreement survives a change of account manager or of platform.

The registration platform changed between the 2025 and 2026 editions. The old export had 61 columns and everybody knew which ones mattered, in the sense that two people knew and one of them has left. The new export has 74 columns, different names, and a country field that arrives as a full name where the old one arrived as a two letter code. Three reports break in the first week and nobody can say what the old feed was supposed to contain, because the agreement was a conversation with an account manager who no longer works there.

Data contracts with registration vendors exist for exactly this moment. A contract is a written, testable statement of what a feed contains and when it lands, and its value shows up on the day the platform underneath it is replaced.

What is a data contract with a registration vendor?

The Open Data Contract Standard, the vendor neutral YAML specification maintained by the Bitol project under the LF AI and Data Foundation and currently at version 3.1.0, puts the definition in one line: a data contract defines the agreement between a data producer and consumers. That framing is the useful part. The producer is your registration platform. The consumers are your loader, your warehouse and every report built on top.

What makes it a contract instead of documentation is that each statement can be checked by a machine on arrival. The country column contains a two letter code is a contract clause. The country data is accurate is a wish.

Sanderson, Freeman and Schmidt's 2025 book on data contracts frames the problem this solves as pipelines breaking and data consumers losing trust, and sets out an architecture with continuous integration, monitoring and version control among its components. For an event team of two or three people, the version control part is the one that does the immediate work, because it is what turns the agreement into something that survives staff turnover.

The four things worth agreeing first

Vendors will discuss a short list and stop reading a long one. Four items carry most of the value.

  • The primary key and its guarantee. Which column identifies a registration, and the promise that it is unique within an event edition and stable across exports. Stability is the part vendors forget, and a key that is regenerated on every export makes incremental loading impossible.
  • The required columns and their types. Not everything they send. The columns you would stop work over, with the type each one arrives as.
  • The arrival window. The file lands by 08:00 local on each day the event is on sale. Say what happens on the day it does not.
  • The notice period for changes. How many days before a column changes shape you will be told, and by what route.

Everything else is negotiable and most of it is unnecessary. The specification above defines eleven sections, including pricing, support channels and service level agreements, which is the right scope for a general standard and more than a registration feed needs in its first version.

How long a notice period do you actually need?

Work it out from your own change cost rather than picking a round number.

Suppose a column rename on the registration feed takes your team twelve working days to absorb: two days on the loader and the manifest, three on the transforms, four across the four reports that reference it, and three on a regression check against the previous edition's numbers. That is the real cost, and most teams have never measured it.

A thirty calendar day notice period is about twenty one working days. Twenty one minus twelve leaves nine working days of slack, which is enough to survive one person being on leave. A ten day notice is seven working days, which puts you five days short before anybody is ill.

The show calendar tightens it further. Those twelve working days do not exist in the six weeks before a show opens, when the same team is running pacing reports daily and rebuilding the on-site dashboards. A contract that says thirty days should also say that no change lands inside a show window, with the windows listed by date, because a vendor cannot know your calendar and will otherwise ship a release on the Monday of your biggest week.

Why the contract outlives the platform

Write the contract against your requirements and it becomes portable. Write it against the incumbent's export and it dies with them.

The difference is concrete. The incumbent sends 61 columns. Your reports reference 18 of them. A contract listing 61 columns is a description of one product. A contract listing the 18, with types and a key, is a specification any platform can be measured against, and it goes straight into the tender document as an appendix. The question you ask each candidate platform stops being what does your export look like and becomes can you produce this.

That reframing is worth more than the contract's enforcement value in most event businesses, because platform changes are the moments when reporting continuity actually breaks. The 2026 edition above lost three reports for a fortnight. A team holding the 18 column specification would have run it against the new platform's sandbox export during procurement and found the country format difference before signing.

Keep the file in version control next to the loader, with the change history visible. Two years later the useful question is when the arrival window moved from 08:00 to 10:00 and who agreed to it, and a commit log answers that in seconds.

What to do when the vendor will not sign

Write it anyway.

An unsigned contract is still the specification your loader asserts against on arrival, and the assertions do not care whether anybody countersigned. The difference is what happens on a breach: with a signature you have a conversation about remedy, and without one you have a conversation about facts, which is still considerably better than the conversation you have now.

The unsigned version also gives you the evidence for the next renewal. Twelve months of arrival times against a stated 08:00 window, with a count of days missed, changes a procurement discussion in a way that no amount of complaining about data quality ever does. Turning the clauses into assertions that run before the load is its own piece of work, covered in K29, and it is the part that produces the evidence.

I would go one step further and put the specification in front of the vendor even when you expect a refusal. Vendors do occasionally agree to a notice period, because it costs them almost nothing, and the ones who refuse tell you something useful about how the relationship will go.

Where a contract has no power

A contract is a document, and documents do not create bargaining power that the commercial relationship does not already contain. A platform holding most of the market in your vertical, integrated with your badge printing and your exhibitor portal, will sign nothing and change what it likes. Pretending otherwise wastes a quarter.

Contracts are also silent about meaning. A clause can require that registration_type is a string from an agreed set of values. Keeping the agreed set current as the show adds a new badge category is a human process, and the day somebody adds a category without telling anyone, every clause passes and the report is still wrong.

The last limit is scope. This is a contract about shape and timing, and it says nothing about what the vendor may do with the personal data flowing through the feed, which is a separate document with separate signatories and belongs to cluster U. Do not let the two merge into one negotiation, because the data protection terms move slowly and the schema clauses need to move fast.

Start with your single most important feed and one page. Write down the key, the columns your reports reference with their types, the arrival window and the notice period you want, and put it in the repository next to the loader. That page is the input to the fuller clause list in K28 and to the manifest that catches a change already made in K22, and it is the first artefact in your unified data work that will still be useful after the platform underneath it changes.

Questions people ask about data contracts with registration vendors

What is a data contract with a registration vendor?
It is a written statement of what a feed will contain and when it will arrive, specific enough that software can check it. Four items carry most of the value: the primary key and its guarantee, the required columns with their types, the window the file must land in, and the notice you get before any of that changes.
Do registration vendors agree to data contracts?
Some will, particularly at renewal or during a tender, and smaller vendors more readily than the largest. Where a vendor will not sign, the document is still worth writing as a one sided specification, because it defines what your loader asserts on arrival and what you hand to the next platform you evaluate.
How much notice should a data contract require for schema changes?
Enough working days to absorb the change before your next reporting deadline. If a rename takes your team twelve working days to work through the loader, transforms, reports and a regression check, a thirty day calendar notice leaves about nine working days of slack, and a ten day notice leaves you short.

Related reading

All data quality articles