Know coverage before the visit.

Every source maps to one result schema you define: the same copay, deductible, out-of-pocket, and network fields whether they came from a 271 or a portal screen-read, with the source recorded on each field so differences stay visible instead of being averaged away. Service-specific detail is pulled for the visit types you configure. When a portal presents a benefit the schema doesn't model, the field comes back flagged rather than silently dropped.

Ambiguity stops the run rather than shipping a guess. An incomplete or internally conflicting benefit answer routes to a person with both sources' raw responses attached, so staff resolve it once with evidence instead of re-running the check. A clean 'not covered' is a successful result, recorded as such. A portal timeout or login failure is a technical error, retried within your rules and then escalated, never blended into the benefit data.

The agent reads the page as it renders on each run, so layout changes are absorbed rather than breaking a recorded path, the failure mode of in-house Selenium builds that go stale weeks after whoever wrote them moves on. If a redesign genuinely removes or relocates a benefit the workflow needs, the run stops with the change identified, and maintenance is Asteroid's job, not a script your team owns.

Checks run on HIPAA-compliant, SOC 2 Type II infrastructure under a BAA, using your portal logins with least-privilege scoping you can rotate or revoke. Each check leaves a per-execution record: what was queried, what came back, the source evidence, and a timestamp. That's the trail that answers 'what did we verify before the visit, and when.' The full audit and access-control posture is documented on the security page.

Regional plans are the reason this category exists, so adding one is routine: scope it against a worked example in the builder, or hand it to Asteroid to run managed. Coverage is chosen by your payer mix, not by what a clearinghouse happens to support. Most deployments add their two or three problem plans first, precisely the ones no 271 covers well.