Your data. Your environment. Your terms.
Zkeleton builds a private environment inside your own VPC where your data finally works for you. Raw PHI never leaves. Core systems are never touched.
Receiving the data is not the same as using it.
The Payer-to-Payer Data Exchange mandate requires every US payer to receive five years of member clinical history by January 1, 2027. Most payers will build pipes to ingest this data for compliance. The data itself remains unstructured, unmatched, and unmerged with claims or authorization context.
A parallel flow in your own VPC.
Ingest from compliance
Zkeleton attaches to your existing CMS-0057 data stream. No new external connections. No new data rights.
Normalize and match
Raw clinical data is normalized against a common schema and matched to your member records with confidence scoring. Low-confidence matches are quarantined for review — never silently merged.
Unify with history
Matched clinical data merges with your claims and prior authorization history into a single dataset inside the bubble.
Payer-owned analysis
The unified data lives inside a bubble in your VPC. Your teams use it for their own analytics. Intelligence leaves the bubble. Raw clinical data does not.
The environment opens
Once the dataset exists, the bubble becomes a place others connect to, on your terms.
What unified clinical data unlocks.
Full clinical context on every claim
Match procedure codes against clinical evidence. Surface billed services without corresponding clinical documentation.
See your at-risk codes before submission
Pre-submission triage flags HCC codes no other source corroborates — or that show a data-entry anomaly — and routes them to your coders with full lineage, before submission.
HEDIS measures against unified member data
Cross-reference clinical history with claims to catch missing screenings, uncontrolled chronic conditions, and follow-up lapses before the measure closes.
Context-aware authorization review
Reviewers see the full longitudinal clinical picture, not just the faxed prior-auth request. Faster approvals where clinical justification is clear.
Payer-owned models on unified data
Run your models against the merged dataset inside the bubble. No data leaves. No vendor cloud retention. Payer policies govern every query.
Use cases expand with the dataset. Anything a payer can build on a unified clinical and claims view.
HARROW: the first workload is already running.
In May 2026, OIG reported that CMS potentially overpaid MA organizations ~$462 million on acute stroke diagnosis codes — in the audited sample, 97 of 97 were unsupported. The failure is structural: an acute-stroke code on a professional claim with no matching facility claim in the same service year. OIG asked CMS to build a prepayment check for exactly this pattern.
That check is HARROW, our pre-submission structural screen, running against the published pattern on fully synthetic data: the uncorroborated code flagged AT-RISK before submission, control cases proving it doesn't cry wolf, and a provenance receipt behind every finding — hashed inputs, versioned rules, reproducible by an auditor. Deterministic rules, no model in the decision path; every uncertain case routes to a certified coder.
demo on request · synthetic data only · no PHI anywhere
Receipts prove lineage and reproducibility — not medical necessity, and not a RADV outcome.
What your environment can host.
Built to shrink your vendor list, not join it: capabilities you currently rent as extracts become workloads you own. Once the bubble exists, it becomes the place others come to work — on your terms, at your pace, with every byte logged.
Opaque extracts become governed exports. Your current stack performs better on higher-fidelity data, and you finally see what leaves.
Vendors bring their compute inside the bubble. Results leave. Data never does.
Device and digital-health makers contribute member-consented data and prove outcomes on your own population before you cover them.
Research that today runs on a broker's copy of your data runs inside your walls instead — you approve the study, queries come in, only aggregates leave.
Four asymmetric choices.
Payer-owned data
The unified dataset is your asset. It resides in your infrastructure under your control. Zkeleton operates on it. Zkeleton does not hold it.
Payer-owned VPC
The entire system runs inside your virtual private cloud. Zkeleton is a guest in your infrastructure, not a landlord holding your data in our cloud.
Payer does nothing new
We operate on the data flow you are already mandated to build. No new operational lift. No changes to adjudication, MLR reporting, or audit trails.
One-sided upside
Upside accrues to the payer: fraud catch, risk accuracy, AI enablement. Transformation cost and architectural burden stay with Zkeleton. The bubble runs beside adjudication, not through it. No critical-systems rewrite at pilot.
We are not building an exchange network. Exchange networks aggregate clinical data through shared pipes in a vendor cloud. Zkeleton runs inside each payer's own VPC. No shared pipes. No vendor cloud. No data leaving payer control.
Reconciled, lineage-complete data — provenance on every merge — is why this works and why it cannot be copied: every alternative architecture pushes complete data into shared or core systems, which is exactly what payers are right to refuse.