Hammer LabsA hospital that does not exist.
A hospital sends a claim, the insurer answers it, money moves to a bank, and somebody writes it into a ledger. Money falls off between those steps and nobody notices for months. Genesee is a whole hospital’s worth of that paper trail, invented from real public numbers, with every place the money fell off written down.
What you get
A zip. Inside it: the claims, the insurer’s replies, the bank deposits, the ledger, the contract, the clinical notes, the coverage rules; sixteen record types, each written the way the party that writes it in life would write it. The answer key: which lines were shorted, lost, denied wrongly or never posted, and why, in the key’s own words. The scoring tool, so you can grade an AI against that key yourself. A manifest with the hash of every file, so a book can be checked and, from its seed, made again to the byte.
The standard book is 2,400 claims over 18 months. Sign in with an email and it is yours, one a day per person. The same seed makes the same book, so two people can work on one copy.
What is real, and what is fabricated
Real, and named inside the zip with where it came from
- The hospital’s population
- payer mix, age bands, sex split, from NY SPARCS and the CMS cost report for the anchor hospital
- The market’s code mix
- every unsuppressed row for Rochester, NY in the CMS Part B file: 3,907 providers, 1.86 million services
- What each insurer publishes it pays
- a commercial payer rate extract built from Transparency in Coverage files
- The Medicare fee schedule
- the 2026 Physician Fee Schedule relative value file at the anchor’s locality
- The table of billing edits
- the published procedure-to-procedure pairs and unit ceilings the insurers enforce
Fabricated, and about nobody
- Every patient
- two hundred of them, with birth dates drawn from the hospital’s own age bands and nothing else about them
- Every claim
- 2,400 over 18 months, 4,081 service lines, from a provider whose NPI is deliberately invalid
- Every payment
- the remittances, the settlements, the bank deposits and the ledger postings, and every place money fell off between them
Every figure the book is shaped by is labelled measured, derived or declared, and nothing is presented as one when it is another. The book represents no real person and derives from none. Cite it as Genesee, then the version and the seed your book carries.
Why it is worth anything
Every other synthetic patient dataset stops at the diagnosis. None of them model what an insurer does with a claim, and that is where the money is lost. Genesee starts where they stop. And because we hid the faults ourselves, we know exactly where they are: on real claims you find out months later, or never, whether the AI was right.
Coming
Hammer Remit
The checks that read those records and say where each claim’s money stopped, what it should have paid against the contract, and what is worth appealing; and the assistant you ask about them. Live today on Genesee, as the demo. For your own records, in your own walls: a conversation.
RCM-Bench
A copy of Genesee we generate with a fresh seed and keep the answers to. Run your AI on it, send us its answers, and see how much it found beside everyone else’s. Once Genesee is out.
A bigger book
A year instead of a quarter, a different hospital or market, many seeds at once, or commercial use. Ask on the download page and a person builds it.
Something wrong with a book? Send it to hello@hammer.ai, with the seed.