Insurance claims execution
Claims intake, demonstrated once, replayed exactly — and checked against the claims database.
See OpenAdapt run the full loop on a real open-source insurance system: a health-facility claim is entered once in openIMIS, compiled into a governed local program, and replayed with a fresh claim number — with success established only by a direct SQL read of the claim row, never by pixels or self-report.
Use supported APIs for adjudication, coverage, and core claims logic. OpenAdapt completes the repeated last-mile intake or update trapped in a portal or claims UI, then verifies the result against the claims database.
Real reference workflow
From demonstration to verified openIMIS claim
Synthetic local data, real openIMIS claims-intake interactions, and an oracle outside the replay path.
- 1.0
Demonstrate
Enter one synthetic health-facility claim — policyholder, claim number, diagnosis, service — while OpenAdapt captures the browser evidence and input events.
Demonstrate — captured openIMIS frames · synthetic local fixture - 2.0
Compile
Turn the demonstration into a parameterized, inspectable workflow with a declared business effect.
workflow.jsoncompiled- workflow
- openimis-claim-intake
- parameters
- insuree no. · claim no. · explanation
- target
- Health Facility Claim form
- effect
- exactly one claim row, status Entered
- healthy run
- deterministic · zero model calls
Compile — inspectable task contract derived from the recorded synthetic workflow - 3.0
Replay
Execute the compiled steps locally against openIMIS with a fresh claim number, without asking a model to reinterpret the task.
Replay — real compiled run · local, model-free, independently checked - 4.0
Verify the write
Check the saved claim through a direct SQL read: exactly one new claim row in status Entered, for the demonstrated policyholder and facility.
independent effect evidenceverified- SQL claim read-backexactly one new claim row
- Claim statusEntered, ready for review
- Policyholder identityinsuree and facility matched
- Executionrecorded save + replays all verified
0 duplicate claims0 wrong-policyholder writes0 model callsVerify — every run is accepted only by a direct SQL read of the claim row in the pinned local reference environment
openIMIS 25.10 (the open-source insurance management system used by national health-insurance schemes) was pinned by image digest in a local fixture loaded with the upstream synthetic demo dataset plus one synthetic policyholder. The clips show one recorded claims-intake demonstration and a real compiled replay on that environment. The evidence manifest records the exact software, task, oracle, media hashes, and scope. OpenAdapt is unaffiliated with the openIMIS Initiative.
Inspect evidence manifestWhy the oracle matters here
The costly failure in claims operations is not a crash — it is the claim that is silently entered twice, or against the wrong policyholder, and surfaces weeks later in reconciliation. Every run in this reference is accepted only when the claims database shows exactly one new claim row, in status Entered, for the demonstrated policyholder and facility. A duplicate or missing row fails the run loudly instead of reporting success.
Where the execution layer fits
- First notice of loss or claims intake re-keyed from an existing structured source into a portal or claims UI with no supported API.
- A bounded status-update or document-attachment step at the edge of an otherwise API-driven claims pipeline.
- Hand QC and compliance an illustrated report of every run: what ran, what it saw, what changed.
Put one claims workflow into production
Bring one repeated intake or update step and the claim record that proves its outcome. We'll map the deployment, verification, shadow run, and supervised rollout.
Book a demo