# Source checks
URL: https://backworkhealth.com/docs/source-checks

When Backwork last fetched a policy, the hash of what it fetched, and the sampled accuracy audits measured for the policy's source

Every policy object in a v1 response has a `source_check`. It tells you where Backwork fetched the policy, when it last fetched it, a hash of what it fetched, and how accurate an audit found the records from that source.

```json
"source_check": {
  "source_url": "https://www.molinahealthcare.com/policies/botox.pdf",
  "last_fetched_at": "2026-09-28T04:00:00.000Z",
  "content_sha256": "9f86d081884c7d659a2feaa0c55ad015a3bf4f1b2b0b822cd15d6c15b0f00a08",
  "source_accuracy": [
    { "field": "overall", "sampled_at": "2026-09-15T00:00:00.000Z", "sample_size": 120, "accuracy": 0.9667, "ci_low": 0.917, "ci_high": 0.987, "method": "manual_review_v1" },
    { "field": "codes", "sampled_at": "2026-09-15T00:00:00.000Z", "sample_size": 120, "accuracy": 0.9333, "ci_low": 0.875, "ci_high": 0.966, "method": "manual_review_v1" }
  ]
}
```

| Field             | Meaning                                                                                                                                                                            |
| ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `source_url`      | The URL Backwork fetches the policy from.                                                                                                                                          |
| `last_fetched_at` | When Backwork last fetched the policy. `null` when no fetch time is recorded. The payer may have changed the document since.                                                       |
| `content_sha256`  | SHA-256 of the fetched content, as 64 lowercase hex characters. `null` when the loader did not record one. Compare it across responses to see whether the fetched content changed. |
| `source_accuracy` | The latest audit of each field for the policy's source, `overall` first. `null` when the policy has no recorded source or the source has no audit yet.                             |

`source_check` itself is `null` only when the policy is no longer on record.

## Sampled accuracy [#sampled-accuracy]

Backwork audits a sample of the policy records from each source against the source documents. Each entry in `source_accuracy` is the most recent audit of one field:

| Field               | Meaning                                                                                        |
| ------------------- | ---------------------------------------------------------------------------------------------- |
| `field`             | `overall` for the record as a whole, or one field: `title`, `effective_date`, `codes`, `link`. |
| `sampled_at`        | When the audit drew its sample.                                                                |
| `sample_size`       | How many records the audit sampled.                                                            |
| `accuracy`          | The share of sampled records that matched the source, from 0 to 1.                             |
| `ci_low`, `ci_high` | The 95% Wilson score confidence interval for `accuracy`.                                       |
| `method`            | How the audit decided whether a record matched.                                                |

The figures describe a sample of the source's records, not the policy in the response. Use the interval, not only the point estimate: with a small `sample_size` the interval is wide. Before you rely on a policy for a coverage, billing, or prior-auth decision, read its source.

The Backwork app and the public policy pages show the same data as `Source checked September 28, 2026` and `Sampled accuracy for this source: 97% (95% CI 92–99%, n=120, September 2026)`.

Code entries carry their own check, `grounding`. See [Code sources](/docs/code-sources#grounding).