Today you read a report. Soon you stop the overpayment before it leaves.
Velora ships in two tracks from the same warehouse. Mode A is what runs in production now — 18 post-service products that turn a claims file into a fiduciary report. Mode B is the next track: real-time pre-payment underwriting that sits in the 837 stream and routes claims to a held-claims queue before adjudication.
One warehouse. Two surfaces. Different buyers.
Reports and real-time aren't sequential phases — they're parallel product tracks for different customer profiles. Brokers, employers, RBP and IDR vendors buy Mode A. TPAs and self-administering employers buy Mode B. Some buy both.
Drop a file. See your audit. Share the report.
Post-service. The customer hands us a claims CSV (or pretokenized via sidecar) and we surface top-N overpayments, fiduciary grade, network shape, leakage, balance-bill exposure. Eighteen products live today, accessible via portal or public API.
- Latency: seconds to minutes
- Verdict: top-N overpayments + fiduciary grade
- Buyer: broker · employer · RBP · IDR · stop-loss
Plug into the adjudicator. Stop the bad claim before payment leaves.
Pre-payment. We sit between the carrier's 837 stream and the TPA's adjudicator, score each claim against the rate floor, hold the outliers in a queue, and route an 835 back with the verdict. The EDI engine is complete; what's left is the held-claims dashboard and first-carrier SFTP handshake.
- Latency: P99 < 500ms warm
- Verdict: GO · WARN · FLAG per claim
- Buyer: mid-market TPA · self-administering employer
Both tracks share the rate warehouse, the PHI vault, the audit log, the API key path, and the compliance posture (HIPAA · SOC 2 readiness · HITRUST e1 · WCAG AA · NIST 800-53).
What changes when the verdict happens before payment.
The Mode A pitch is forensic — "look what your TPA paid above the published rate last quarter." The Mode B pitch is structural: the overpayment never lands. Velora's score sits in-line on the adjudication path, holds the outliers, and routes them to a queue your team triages in minutes, not after-the-fact.
837 in
Carrier streams the live claim file in over SFTP, AS2, or REST. Velora parses, normalizes, and runs the rate-fairness score against the contracted MRF rate for that NPI × CPT × carrier.
GO · WARN · FLAG
Each claim gets a verdict. GO claims continue to adjudication untouched. WARN and FLAG claims land in the held-claims queue with the rate reference, the variance, and a suggested action.
835 out
Once the queue clears, an 835 routes back to the carrier with the original payment OR the adjusted amount + reason codes. Audit log captures the full chain, retention 7 years.
What's built. What we owe.
Mode B is 94% built today. The remaining work is operator-shaped and customer-shaped, not engineering-shaped. Both columns shipped honestly so prospects know what they're signing up for.
Built
Shipped · in production- EDI engineAll X12 5010 transactions implemented and tested — 270/271, 276/277, 277CA, 278, 834, 835, 837P/I, 820, 999. Forty carrier profiles. 4,200+ tests passing.
- Pre-claim score endpointLive at
/api/v2/preclaim/score— the rate-fairness scoring that powers each verdict. - Tenant + audit infrastructurePer-tenant rate limiting, customer API keys, signed webhooks, 7-year audit retention. Production-grade.
Owed
In active development- Held-claims operator dashboardThe queue UI where TPA staff triage WARN/FLAG claims. Scaffold landed; finishing the workflow with our first design partner.
- First carrier SFTP handshakeProduction cutover happens with the first design-partner TPA — that's the only way to verify the real-carrier handshake holds.
- Per-tenant PHI scrubber + BAAsFour BAAs in legal review (ClickHouse, Neon, Railway, R2). Single shared blocker before either mode handles production customer data.
Pilot Mode B with us.
We're taking two design-partner TPAs for the real-time track. White-glove onboarding, your held-claims queue lands on our roadmap. The trade: we get to learn against real 837 traffic before general availability, and your name lands on the launch case study (if you want it).