Skip to content
← BLUEHEALTHCARE
blueClaims

blueClaims

Claims shadow-review, health and P&C lanes

Anchors every finding to the exact policy clause behind it, and to the specific record document and its hash. Types every “not found” so a denial can never claim absence of evidence the review couldn’t have looked for.

Flagship — demo-ready
Runs locally · CLI or web UI · Ed25519-signed receipts · no cloud account required
Who it's for

Claims adjusters and compliance leads at a payer or carrier.

What it is not

blueClaims never approves, denies, pays, reserves, or alters a claim. The human adjuster is the only adjudication authority.

Clinical text stays where it is. A record anchor names the document and its hash, not a quoted passage. Capturing an excerpt into a packet is a separate step your clinic turns on and approves — off by default, so no clinical text leaves a record unless someone decided it should.

How it works
1
Load a claim, its policy, and its supporting records — synthetic in the demo, real files in a pilot.
2
The engine maps every clause to its evidence and surfaces contradictions and missing evidence.
3
Each gap is typed SEARCHED-ABSENT, NEVER-SEARCHED, or UNDETECTABLE-WITH-THESE-RECORDS — never a single undifferentiated “not found.”
4
The adjuster reviews the packet in the Workbench, reaches a decision, and exports it in one of five formats.
5
The signed packet is independently verifiable offline.
Screens

blueClaims works a claim in shadow: the policy, the record bundle and the claim's service lines are loaded together, every finding is tied to the plan clause and the record document (by hash) that earned it, and every "not found" is typed so nobody can later claim absence of evidence the review could not have looked for. The screens below follow the bundled synthetic health claim from an empty library to the signed receipt.

blueClaims Claim review library screen in its first-run state: the suite shell's sidebar (Case intake with Library and New analysis, Evidence map, Reviewer, Case file) beside an empty table reading No claims yet, a New claim button, and three starting cards — Start from files, Open sample, and Open health sample.
Synthetic demo data
First run — nothing is loaded until you load it. Three ways in: drop a complete source bundle and let the preflight classify it, open the property-loss sample, or open the health-claims sample.

Opening both samples gives the library below.

blueClaims Claim review library screen: the suite shell's sidebar (Case intake with Library and New analysis, Evidence map, Reviewer, Case file) beside a table of two synthetic claims — a health claim with three service lines marked Cannot certify, and a property-loss claim marked Needs human review — with Import prior work and New claim buttons and a Property claims toggle in the top bar.
Synthetic demo data
Claim review library — one health claim and one property-and-casualty claim, each with its policy version, record bundle, evidence packet and adjuster review kept together. Both states shown here mean the same thing: an adjuster has to look.

Open lands on the packet.

blueClaims Claim evidence packet screen: five count cards — 3 service lines all mapped, 1 Supported dual-anchor, 1 Discrepant, 0 Ambiguous, packet CANNOT-CERTIFY adjuster required — above two findings requiring attention: lab-panel-80053 Discrepant, the billed amount of USD 420.00 against a recorded fee-schedule amount of USD 340.00, and arthroscopy-29881 Cannot certify because its finding class was not addressed by this run's mapping; a Result readiness panel lists the claim, plan and contract identifiers, and the subtitle states that no payment or coverage disposition is produced.
Synthetic demo data
Claim evidence packet — the two findings an adjuster must look at, in words: a billed amount that does not match the recorded fee schedule, and a service line the mapping never addressed. The packet is CANNOT-CERTIFY because of the second, not because anyone judged the claim. Re-run, Inspect evidence and Export are the only actions.

Inspect evidence shows what each finding is tied to.

blueClaims Plan and clinical document anchors screen: three claim objects with verdicts Supported, Discrepant, and Cannot certify, with an evidence inspector showing the exact plan clause and clinical document that anchor a Supported verdict.
Synthetic demo data
Plan and clinical document anchors — office-visit-99213 is Supported by plan provision 3.1 and the office-visit note, each named with its SHA-256; lab-panel-80053 is Discrepant against provision 4.5 and the lab report; arthroscopy-29881 has no plan anchor at all. The inspector's confidence line reads Not inferred — deterministic only.

Only the document name and hash are recorded here; the clinical text itself stays in the record unless the clinic has turned excerpt capture on. Export writes the signed receipt below.

blueClaims Signed process receipt screen: a synthetic-data warning that no real claimant, member, or patient information was processed, a Run offline verification button, and a receipt body naming the property-loss claim, its policy id, state NEEDS-HUMAN-REVIEW, and a wind-damage-roof finding with verdict SUPPORTED.
Synthetic demo data
Signed process receipt — this run is the property-loss claim, state NEEDS-HUMAN-REVIEW, with its input hashes and payload hash listed above the finding-by-finding body. Run offline verification performs the identical check without an account, a product database, or a MOBLUEHQ server.
Status, measured

Engine, receipts, persistence, CLI, API, and a working web UI are real and tested for both lanes (health and property & casualty). All five export formats, run ancestry, interrupted-run recovery, drafts/autosave, seeded-scale performance testing, an accessibility audit, and clean-machine packaging are done.

Not yet true: Library pagination at large record counts is a known, named gap. No signed pilot. No live PHI.

A signed receipt proves what blueClaims ran and recorded; it does not prove an underlying real-world claim is true. blueClaims does not approve, deny, recommend, pay, retract, or alter anything a named human is responsible for deciding.