Policy thresholds

Use Policy Thresholds to set the reporting levels that need attention and the response to selected policy events. These settings apply across the organization. Agree who will investigate a reported miss before changing them.

Set reporting thresholds

Set the minimum acceptable percentage for:

• Human oversight recorded on decision steps.
• Model steps with a supporting source, shown as Grounding Rate.
• Safety checks that pass, shown as Guardrail Pass Rate.
• Screened responses without a content-safety flag.

VeriProof flags a measure when it falls below its threshold. The same values appear in governance reporting and policy summary email.

For a 90% threshold, a recorded rate of 85% is below the required level. Lowering the threshold to 80% changes which results are flagged; it does not improve the recorded coverage.

Choose policy notifications

Enable content-safety incident notifications when responders need individual notices. Use the weekly policy summary for a regular view of coverage and measures below their thresholds.

Notification delivery still depends on the organization channels and audiences configured under Settings → Integrations → Notification Channels.

Require oversight for decision types

Select each decision type that must include a recorded human oversight event. Sessions that do not meet the requirement appear in coverage-gap reporting.

A reporting requirement is not proof that a person reviewed the decision. Check the session’s recorded oversight evidence when investigating a gap.

Set content-safety responses

For each supported content-safety category, choose the saved response:

• Default follows the Content safety incident alerts setting.
• Log only records the flag without creating an alert or blocking the event.
• Alert records the flag and creates the configured notification.
• Block selects blocking as the required response.

Make a controlled change