This screen shows whether the input to a model in operation has changed from the training data (drift). The CLI (geo-mlops-client drift evaluate) or a batch job uploads the measurements with the operate permission. The server decides on drift using the thresholds of each model's policy. Every member sees it; saving a policy needs settings (ADMIN).

  • PSI (population stability index): how much the current distribution differs from the baseline distribution. At or below 0.1 they are nearly the same; above 0.2 the change is large.
  • KS: a statistical test of whether two distributions are the same. If the p-value is below the threshold (default 0.05), the distributions differ.

How drift is judged: a CLI or batch job compares the baseline with live input, computes PSI and KS and sends them to the central server. The server judges them with the per-model policy and, on drift, records a WARNING alert and a CT trigger. The CT trigger never starts training; the AI developer presses Retrain this model on the drift screen, which opens an empty new training wizard

Drift list

Drift — ① search and status filter ② PSI ③ KS ④ status
#AreaDescription
①Search and filterSearch models (모델 검색), status (All · Drift · Normal (전체 · 드리프트 · 정상)).
②PSI (PSI)PSI of the last measurement.
③KS (KS)KS statistic of the last measurement. The p-value used for the verdict is not shown on screen.
④Status (상태)Verdict according to the policy in effect when the measurement was uploaded. Changing the policy does not change verdicts already recorded.

Each row is the last measurement of one model. The last measured (마지막 측정) column shows relative time; click the model, PSI or last measured header to sort. Click a row to open the per-model detail.

Per-model detail

Drift detail — ① retrain this model ② PSI trend (30 days) ③ CT trigger history ④ drift policy
#AreaDescription
①Retrain this model (이 모델 다시 학습하기)Opens the new training wizard (when the training feature is on and you have develop). The wizard opens empty; the model and dataset are not pre-filled.
②PSI trend (30 days)One PSI bar per measurement. At or above the policy's warning threshold they are yellow, at or above the critical threshold red, and both thresholds are drawn as dashed lines.
③CT trigger history (CT 트리거 이력)Retraining (CT) triggers recorded on detected drift. Each row shows a result badge, the cause (drift psi=…) and, if present, the start of the pipeline run id.
④Drift policy (드리프트 정책)This model's thresholds. ADMIN can change them and Save (저장); other roles only see the values. Leaving the page with unsaved changes asks for confirmation.

Click the model name in the title to open the model in the model registry.

Policy items:

ItemDefaultMeaning
PSI warning threshold (PSI 경고 임계)0.1At or above this, yellow in the trend. Not used for the verdict
PSI critical threshold (PSI 위험 임계)0.2At or above this, Drift (드리프트)
KS p-value threshold (KS p-value 임계)0.05If the p-value is below this, Drift (드리프트)
Require human approval (disable auto promotion) (사람 승인 필요(자동 승격 비활성))OnThe value is saved, but there is no automatic promotion yet, so it has no effect. Production promotion always goes through an approval request

The default policy applies until you save. Saving creates a policy for this model only.

Statuses and badges

BadgeValueMeaning
Normal (정상)NORMALPSI is below the critical threshold and the KS p-value is at or above the threshold
Drift (드리프트)DRIFTPSI ≥ critical threshold, or KS p-value < threshold. On this verdict a WARNING alert is created and a CT trigger is recorded.

Written for the platform as of 2026-10-02.

© MLOps Central Platform