Essay
Why audit scores drift after the first quarter
Launch week of a scoring model often looks tidy. Definitions are fresh, reviewers are attentive, and leadership is still reading the model memo. By the third month, small exceptions accumulate. The risk scoring audit platform still produces numbers — they simply no longer mean what the memo claimed.
Threshold creep
Someone softens a band edge to avoid an awkward conversation. The change is informal. Next quarter another softening follows. Without a change log tied to criteria, the published scale becomes a negotiation history rather than a standard.
Silent overrides
Overrides are sometimes necessary. The failure mode is an override without a reason code, expiry, or owner. We teach an override log that treats exceptions as temporary, visible, and reviewable — not as a back door that rewrites the model.
Staffing without handover
When a calibrated reviewer leaves, their private heuristics leave with them. Criteria sheets with non-examples reduce that loss. Pairing a new reviewer with a short calibration sample is cheaper than discovering divergence during an external audit.
Stewardship beats relaunch
Teams often respond to drift by redesigning everything. A lighter ritual works better: quarterly drift checks against a frozen sample, a short note on criteria that moved, and a refreshed memo paragraph for stakeholders. Module seven of Defensible Risk Scoring is devoted to this hygiene.
Drift is not a moral failing. It is what happens when a living system has no steward. Name the steward, schedule the check, and keep the log boring — boring is the point.