An ATS audit should show what the assistant intended to write, what it attempted, and what the destination actually stored. A call summary saying “updated successfully” is not enough evidence.
Define an audit record for each action: candidate context, application where relevant, source call reference, approved field scope, intended value, operation time, actor or policy, remote identifier, and result. Keep sensitive content restricted under policy.
Verify the destination's current semantics
Greenhouse's note-creation reference distinguishes note types and candidate versus application anchoring. Review the current documentation for the actual ATS operation instead of assuming every readable entry can be created or edited the same way.
Grant only the write permissions the administrative workflow needs. A prompt restriction cannot fully compensate for broad credentials and an unconstrained action interface.
Work through a fictional mislinked note
A candidate discusses interview preferences for one role while holding two applications. The intended action is an application-specific scheduling note. The audit compares the selected candidate and application with the resulting remote record.
If the note lands at the wrong scope, staff receive a correction task with the original source and affected records. The audit should not hide the defect because the API returned success.
| Audit check | Required evidence |
|---|---|
| Correct candidate | Verified record relationship |
| Correct application | Intended vacancy linkage |
| Faithful content | Source meaning and qualifiers preserved |
| Allowed fields only | No unrelated status changes |
| Unknown response | Reconciliation before retry |
Keep partial outcomes visible
A contact update may succeed while a coordinator task fails. Track each action separately and group them under the call for review. Avoid treating the whole interaction as atomic unless the supported implementation actually provides that guarantee.
Detailed audits add storage and review work. A minimal action ledger can preserve enough evidence without copying complete transcripts everywhere. Use controlled supporting records when deeper content review is necessary.
Sample both ordinary successful writes and exceptions. Review faithful summaries and ATS integration boundaries for the upstream rules being audited.
Questions about ATS audits
Does an HTTP success prove the update is correct?
It proves technical acceptance under that operation's semantics. Inspect the resulting record and business scope.
Should every field value appear in general logs?
No. Keep the minimum evidence and apply restricted access to sensitive content under policy.
What should happen after a wrong write?
Preserve evidence, assign a correction owner, reconcile affected downstream work, and verify the repair. Do not silently erase the incident history.