The disparity is the data.

Every year, the CDC publishes a number. The 2024 release said 22.3 maternal deaths per 100,000 live births. Newspapers wrote it down. Health-system PR offices wrote it down. The number has been falling slightly for two years, so the headlines were cautiously optimistic.

Inside that single number live two countries. One has a maternal mortality rate comparable to Germany. The other has a rate comparable to nations with a tenth of our per-capita health spend. The two countries occupy the same hospitals. They share the same OBs. The line between them is the race of the patient.

"A national average is a kind of mercy extended to the system being averaged."

I started SEIAI because I could not let go of a small, ugly observation: the only people for whom the disparity matters are the people inside it. The system has organized itself, almost elegantly, to never have to look directly at what it is doing. National dashboards aggregate. Hospital scorecards risk-adjust. Insurance benchmarks compare to peer groups, which means comparing to systems with the same blind spots.

Aggregation is, in this domain, a form of erasure. You cannot improve what you refuse to stratify.

What stratification actually requires

It is fashionable to say a system is "data-driven." Almost every health system in the United States is data-driven. They drive their data into a warehouse, where it sits next to other data, where it is queried by analysts who produce dashboards for executives who never see the patients. The data drives nothing.

Stratification is the opposite of that. It means: every analytical surface, by default, breaks the population into the axes that matter — race, age, parity, payer, geography, language, gestational age — and refuses to display an aggregate without an explicit override. It means a hospital VP cannot look at their cesarean rate without also looking at their cesarean rate for Black patients, for Medicaid patients, for primigravida patients. The default view is the disaggregated view.

When I describe this design choice to clinicians, they nod. When I describe it to executives, they get quiet. The aggregate has been protecting them. Removing the aggregate is removing a kind of comfort.

Consent is not a checkbox

The other thing aggregation has been protecting is the patient — at least nominally. The argument goes: if we publish stratified outcomes, we risk identifying individuals. So we don't publish. So we don't measure. So nothing changes. Everyone is safe and no one is well.

Consent-native infrastructure dissolves that trade-off. Every record in Genesis Meridian carries four explicit consent axes: care, quality improvement, aggregate benchmarking, public dashboard. The patient decides, at the granularity of each axis, where their record participates. Revocation propagates within 24 hours. K-anonymity (≥11) protects the published surface — and the disparity, importantly, survives the privacy filter, because the disparity is structural, not individual.

"Patients should be able to say yes to their midwife and no to a benchmarking consortium in the same breath, and have the record honor both."

This is not a feature. It is the architecture. It is what the database knows before any application is written on top of it. We did not bolt consent onto a record system. We started with consent and let the record system grow around it.

Why we started with midwives

Independent midwives are roughly 8% of US births and absorb most of the planned out-of-hospital care. They are the smallest practices in the perinatal landscape and the most disciplined about outcomes — they have to be, because they are watched the most closely. They were also the most chronically underserved by EHRs, because no enterprise vendor finds it economical to build for them.

We started there because the wedge into a measurement reform is the population that already wants to measure honestly. The hospital systems will come, eventually, when the public dashboard makes it untenable to continue not coming. But the public dashboard requires data. The data requires practices willing to publish. The practices willing to publish needed software. And the software needed to be designed by people who actually believe that the disparity is the data.

What we are accountable to

The same dashboard that we show investors is the dashboard the public can read. There is no private dashboard with prettier numbers. There is no internal-only metric that contradicts the published one. If our customers' aggregated outcomes worsen, we publish the worsening. If a state's hospital system shows a widening Black:White SMM ratio, we publish that. If our own product has a quarter where adoption stalls, the public dashboard reflects it.

This is not a marketing posture. It is the only way the design works. The whole point of a consent-native, stratification-by-default system is to remove the operator's ability to pick which numbers go to which audience. Removing that ability is the product.

The next forty years

The disparity has been documented since the early 1980s. Forty years of measurement, forty years of unchanged outcomes. We do not believe a software product solves a structural injustice. We believe a software product can change what is visible, what is comparable, what is consented to, and what is impossible to look away from.

That is a small thing. It is also the only thing that has not been tried.