MERIDIAN SPECIALTY HOSPITAL - BASELINE SNAPSHOT
Synthetic source pack. Snapshot date 2026-09-09. Currency USD. Generator seed 20260909.

WHAT THIS IS
A fictional multi-specialty hospital's historical claim portfolio: 5,200 claims across nine
specialties and seven payers, roughly twenty four months of service dates. Upload the whole folder,
or its ZIP, into an empty workspace.

WHAT IS DELIBERATELY ABSENT
No source file in this folder contains a denial family, an expected allowed amount, a variance
against contract, a recovery probability, an expected value, a filing or appeal deadline, a route
recommendation or a root cause. A real 837 contains none of those, and shipping them would mean a
platform was reading an answer key instead of deriving anything. Every one has to be derived here.

THE FOLDERS, AND WHAT REAL SYSTEM EACH ONE STANDS FOR
  claims/              837 shaped claim export from the billing system, in five encodings plus one raw X12 837.
  claim-lines/         Service line detail. Line charges reconcile to the claim charge.
  remittances/         835 shaped remittance export, plus provider level adjustments, plus one raw X12 835.
  adjustments/         Full CAS segment detail: group code, reason code, amount, remark codes.
  acknowledgements/    277CA acceptance archive from the clearinghouse. Perishable, see policies/.
  contracts/           Fee schedule exhibits: professional rates, carve-outs and layered methodologies.
  invoices/            Retained supplier invoices for implant and device carve-outs.
  work-history/        Accounts receivable follow-up notes and the write-off register, with authors and timestamps.
  correspondence/      Denial notices, records requests, status responses, overpayment and policy notices.
  clinical/            Operative notes, imaging reports, procedure reports, therapy records.
  pm-export/           The same remittance cycle as a practice management system exports it after posting.
  inventory/           Specialty catalog of devices, drugs and priced components.
  policies/            Payer companion guide semantics, acknowledgement retention, review boundaries.
  history/            Disjoint historical outcome rows for local model training. No overlap with this portfolio.
  organization.json    Organization context and the payer profiles, with provenance on each behavioural figure.

THE JOINS
  claim_id joins claims, service lines, remittances, acknowledgements, work history, correspondence
  and clinical records. Unstructured documents carry the claim number in the filename and in the text.
  payer plus code plus service date joins a claim line to the fee schedule exhibit in force that day.
  invoice_reference joins a carve-out line to its retained supplier invoice.

  A PROVIDER LEVEL ADJUSTMENT CARRIES NO CLAIM NUMBER. Real ones do not. The adjustment identifier
  field carries the trace of a PRIOR remittance, so a takeback is only discoverable by joining the
  current remittance file to a historical one across time. That join is the whole point and this
  folder does not shortcut it.

THE FOUR POPULATIONS, AND HOW EACH ONE IS FOUND
  Denied        The remittance carries a payer initiated or contractual obligation reason code and
                little or no payment. Fully denied lines pay 0.00 and therefore post no transaction
                at all, which is how posted-transaction denial reports go blind.
  Underpaid     Paid, posted and closed, but below the contracted rate. Invisible in the remittance
                alone. Found only by joining the paid amount to the fee schedule for the service date.
  Stalled       Accepted by the payer and never adjudicated, clock running. Distinct from a claim the
                payer never received, which needs resubmission rather than appeal. The 277CA archive
                is the only thing that tells those two apart.
  Recouped      A prior payment clawed back through a provider level adjustment. Found only by the
                two file trace described above.

DELIBERATE DATA DEFECTS
  DEFECT-01-DUPLICATE-CLAIM (MER-000001): MER-000001 appears twice, byte identical, inside claims/01-claims-institutional.csv. The pipeline must suppress the repeat and keep one case rather than double counting the charge.
  DEFECT-02-MALFORMED-ROW (MER-MALFORMED-ROW): MER-MALFORMED-ROW is the last row of claims/01-claims-institutional.csv. It carries a negative billed charge of -412.00, an impossible service date of 2026-13-45 and an empty submission date. The row must be rejected with its filename and row number, and must not reduce any total. The column count is intact on purpose, so the rest of the file still parses.
  DEFECT-03-NO-REMITTANCE (MER-000642): MER-000642 was submitted and adjudicated in the source A/R system but no 835 line was exported for it. The pipeline must report a missing baseline payer payment rather than assume zero.
  DEFECT-04-UNTRACEABLE-TAKEBACK: A PLB FB offset of 18427.00 USD on trace 877261385F6ENZEM dated 2026-08-17 carries free text in the adjustment identifier field instead of a prior remittance trace. It cannot be joined to a claim in either direction. The pipeline must report it as an unmatched remittance and must not attach it to a guessed claim. Every other PLB row in this folder is traceable, but only through a two file join from PLB03-2 to a historical 835 trace.
  DEFECT-05-FILING-WINDOW-CLOSED (MER-004048): MER-004048 carries a Northwind Medicare Advantage denial notice dated 2024-12-11. With five days presumed receipt and the 65 day appeal window recorded in the fee-schedule exhibit, the appeal right closed before the snapshot date. Expected value is zero however large the balance.

HONESTY
Every identity, payer, plan, amount, rule, note and clinical sentence in this folder is fictional.
There is no protected health information here. Verified incremental cash starts at zero and stays
there until a finance reviewer attributes it. The payer behaviour parameters are anchored to
published figures where published figures exist, and are labelled modelled where they do not.
See docs/SYNTHETIC_DATA.md in the repository for the provenance of every distribution.
