About
We read the federal record so you don’t have to.
Iron Horse Analytics exists because the public healthcare record is genuinely public and genuinely unusable. The files are free. Making them agree with each other is the work — and it is the same work, repeated, in every consulting engagement that touches this data. So we did it once, properly, and built on top of it.
- What we hold
- 16 federal source files, reconciled into one provider record
- How it's checked
- 221 automated checks gate every load
- Privacy posture
- Public data only — nothing HIPAA-covered
The standard
Nothing gets published that hasn’t passed the gate.
Data work fails quietly. A source file changes shape, a load runs half way, a metric drifts — and nobody notices until a client builds a recommendation on it. Most of our engineering goes into making that impossible.
221
Automated checks on every load
pytest
16
Federal source files reconciled
load_manifest
100%
Match against the CMS reference file
provider_metrics
Nothing loads before it validates
Every run opens with a hard gate. A source that arrives truncated, empty, or as an HTML error page dressed up as a CSV fails the gate, and the run halts rather than publishing partial data. We test that by deliberately corrupting inputs and confirming the pipeline refuses them.
Schema drift is caught, not absorbed
Federal files change shape between releases — a column renamed, a code list extended, a form revision that moves a line. The gate compares each incoming file against its expected schema and stops on a mismatch, so a silent change never becomes a wrong number downstream.
Every build is reproducible
Each transformation is a pure function: raw file in, clean file out, same result every time. Re-running the pipeline on the same inputs produces byte-identical outputs, which is what makes an audit trail worth anything.
Derived metrics are checked against CMS
The derived financial layer is validated row by row against the published CMS reference file. Not a sample — every record. When our number and theirs disagree, that's a bug, and it blocks the release.
Who we work with
Mostly people who could do it themselves.
Our clients are consultants, analysts, and strategy teams who understand this data well enough to know how long it takes. They hire us for the linkage layer and the cost report expertise, and keep the analysis in-house where it belongs.
Public data, handled properly.
Every source we hold is a public federal file. Nothing we touch is HIPAA-covered, which means an engagement with us doesn’t have to clear a privacy review before it can start.