What happens when an eligibility file is wrong

A file can parse and still be wrong

Most loaders check format: the right columns and types, a file that opens. That's format accuracy. Content accuracy asks whether the values are true.

A file can pass every schema check and still terminate 400 employees by mistake, or end a member's coverage before it started.

You usually hear about a content error when a member tries to use a benefit and is told they aren't covered. Then it comes back as a complaint to an employer who's deciding whether to renew.

Where a wrong file gets stopped

Every record goes through 5 layers of checks before it moves:

• Type and range for each field.
• Duplicates within the file and against existing members.
• Consistency across a member's records, such as dependents and coverage dates.
• Deviation from the prior file from the same source.
• A change threshold: what percent of the new file differs from the last one, to catch mass changes before they land.

When a check fails

• A record-level failure stops that record. The rest of the file keeps moving.
• A file-level failure stops the file.
• AI proposes the most likely cause. A person on Dart's team makes the call and contacts the sender when the fix belongs to them.
• If a decision holds up a delivery, we tell your downstream partners.
• You can see each flag and the reasoning behind the decision on it.
• We add each correction to that source's validation set, so the same error doesn't get through twice.

AI proposes. A person decides. The file waits.

Over 99% identity resolution, checked by people

Identity resolution runs above 99% across eligibility, claims, provider visits, and benefits data, on about 3 million records a month. Any match below the confidence bar goes to a person on Dart's team. See matching members without a shared ID.