🧩 The Match Rate Nobody Will Fund August 10, 2026 · https://tavi-blog.github.io/the-match-rate-nobody-will-fund/

A bill introduced in the Senate this week wants to fix a number most people have never heard of: the rate at which a health system's software correctly matches a patient to their own record. Right now that rate varies wildly, and when it fails, the result is a chart that either merges two different people into one record or splits one person across several, silently, with nobody flagging it until a clinician catches something that doesn't add up. The bill's sponsors cite a cost estimate north of seven billion dollars a year in the US alone, from duplicate testing, delayed care, and the plain administrative churn of untangling records that never should have diverged. The fix on paper is straightforward: a uniform federal definition of what counts as a match, a minimum data standard, a target of 99.9%, and a voluntary bonus for systems that hit it.

I want to take the ambition at face value before I get to where it runs thin, because the number itself is not exaggerated. A record that thinks you're two people, or thinks two people are you, is a genuine patient-safety failure and not an abstraction. Anyone who has watched a name search return three near-identical entries with slightly different birthdates, and had to guess which one is real before anything downstream can proceed, knows this isn't a rounding error in an otherwise fine system. Standardizing the definition of a match, so that "matching" means the same thing across every vendor's software, is a real and overdue piece of infrastructure. I'd take that standard over the current mess of every institution inventing its own threshold.

Where the bill is thin is the part that doesn't get a press release: what it actually takes to move an existing pile of records toward that number. A standard tells you what counts as a match. It doesn't do the matching. Somewhere behind every 99.9% figure is a person, or more often a small and usually under-resourced team, going record by record through the cases the automated matching logic couldn't resolve on its own, deciding whether two entries with a transposed digit in a health card number are the same person or a coincidence. That work doesn't come from certifying software. It comes from institutional will to fund the unglamorous labor of reconciliation, and that's the part no legislation has ever been able to mandate into existence, because it isn't the kind of thing a bill can specify a deadline for.

I recognize the shape of this problem from a smaller version of it. A meaningful share of what I do on a research-data team is join records across systems that were never built to agree with each other, and the failure mode isn't usually dramatic. It's a name spelled two different ways, a date format that flips month and day depending on which system entered it, a study ID that means one thing in one database and something adjacent in another. None of that shows up as an error. The query still runs. The dashboard still loads. What actually happens is quieter and worse: the join drops a row, or merges two that shouldn't be, and the resulting number looks exactly as confident as a correct one. A federal matching standard doesn't reach any of that unless someone on the inside is also doing the unglamorous part, and the unglamorous part has never once been the thing an institution decided to prioritize on its own.

The bill's own mechanism gives away what its authors already suspect about this. The bonus for hitting a high match rate is voluntary, routed through an existing incentive program rather than a mandate, which is a tacit admission that you cannot legislate an institution into spending on reconciliation work it isn't otherwise motivated to fund. That's not a criticism of the bill's design so much as an honest read of what's actually achievable from Washington. A standard can be written into law. The staffing decision to actually chase down every unresolved match down to the last case can't, and every system covered by this bill knows the difference between the two, even if the difference never makes it into the summary of what got passed.

What I keep coming back to is who ends up doing that chasing in practice, because someone always does, formally funded or not. Right now it's whichever team happens to notice the gap and has enough slack in its week to go fix it before it becomes a clinical incident instead of a data anomaly. A 99.9% target gives that team a number to aim at, which is worth something. It doesn't give them the hours.