18. SafeGuard

SafeGuard is Klariton’s quality monitor for published BIQs. It checks whether live answers still match source material and Brand Voice rules. It flags findings; it does not silently change content.

What SafeGuard Checks

Finding type	Meaning
Claim contradicts	The BIQ says something that conflicts with source material.
Claim unsupported	The BIQ makes a claim that source material does not support.
Framing rule violation	The BIQ violates a Brand Voice or compliance rule.

Severities:

Severity	Meaning
Critical	High-risk contradiction or compliance issue.
Warn	Should be reviewed before relying on the BIQ.
Info	Lower-risk issue or useful quality signal.

Statuses:

Status	Meaning
Open	Needs triage.
In re-review	Sent back to the curation workflow.
Resolved	Fixed or no longer present.
False positive	Reviewed and intentionally closed.

Where To Find It

Open SafeGuard from the dashboard or the Quality Monitor route. The dashboard can also show SafeGuard health when open findings exist.

The SafeGuard page includes:

drift rate,
BIQs with findings,
open findings by severity,
resolved findings in the last 30 days,
filters by type, severity, status, lifecycle, and touchpoint,
sorting by severity, newest, or confidence,
finding detail with BIQ text, material support, judge assessment, and context.

Manual Scan

Use Check now to scan changed published BIQs against source material. The scan creates or updates findings and reports how many BIQs were checked.

Manual scans are useful after:

material imports,
product catalog changes,
Brand Voice changes,
SafeGuard rewrites,
bulk BIQ generation,
release preparation.

Automatic Scan

Automatic SafeGuard scans are configured under Settings -> Brand Voice & Compliance.

Options include:

enabled or disabled,
daily or weekly cadence,
check at BIQ creation where available.

Auto-scan is opt-in and defaults to disabled. A scan can create findings, but the user still decides what to change.

Triage Actions

For each finding you can:

send the BIQ to re-review,
mark the finding as false positive,
open the affected BIQ or material,
generate an AI rewrite suggestion,
edit the suggestion before approving,
fix all findings on the same BIQ,
apply bulk actions to identical findings,
auto-correct all affected BIQs in a selected touchpoint where permitted.

Approval updates the BIQ. After a correction, check translations because source changes can make translated answers stale.

Recommended Workflow

Filter to critical open findings.
Open each detail and compare BIQ answer with source material.
Decide whether the issue is real.
For real issues, send to re-review or generate a rewrite suggestion.
Review the corrected answer before approval.
Re-run translations if needed.
Run Test Center when chat behavior may be affected.

What SafeGuard Is Not

SafeGuard is not a replacement for human review. It detects likely contradictions and rule violations, but product owners remain responsible for final wording and publication.