Model Performance Watch Game

Watch model metrics change, detect hidden failures, investigate drift, manage alerts, compare versions, and protect production reliability through fast strategic decisions in every mission.

Mission control

Configure the production environment, monitoring window, and game rules.
Session not started

Alert thresholds

Tune monitoring sensitivity.

Accessibility

Live model health

Start the simulation to begin monitoring.
Idle
Accuracy
No data
Precision
No data
Recall
No data
F1 score
No data
Error rate
No data
Drift score
No data
Latency
No data
Confidence
No data
Version: v3.2 production
Throughput:
Active alerts: 0
Review queue: 0
Specificity:
FPR:
FNR:
Log loss:
MAE:
MSE:
RMSE:
R²:

Production stream canvas

Animated requests show predictions, errors, drift, alerts, and infrastructure pressure.
Incoming Correct Error Drift
Ingest
Preprocess
Predict
Evaluate
Respond

Performance analytics

Plotly charts update from the selected horizon and monitoring window.

Alerts and investigations

Inspect breached thresholds and likely causes.
No alerts yet.

Select a cause and investigate after an incident appears.

Mission score

Total score
0
Monitoring Beginner
Performance
0
Detection
0
Decisions
0
Reliability
0
Business cost$0
Avoided failures0
Missed warnings0

Corrective actions

Every response changes cost, risk, downtime, and future performance.

Human review queue

Low-confidence and high-risk predictions await review.
The queue is empty.
Review accuracy: 96%
Cost/item: $2
Delay: 0 min

Event log

Metrics, incidents, deployments, decisions, and recoveries.
No events recorded.

Custom monitoring data

Paste labels, predictions, confidence, latency, version, and timestamp records.

Monitoring knowledge challenge

Answer questions about metrics, thresholds, drift, rollback, and retraining.
Quiz score: 0
Start the simulation or load a quiz question.

Incident summary and learning guide

Review formulas, interpretation, limitations, and recommended responses.

Current round summary

No round data is available yet.

Recommended response

Begin monitoring to receive a recommendation.

Core formulas

Accuracy = (TP + TN) / (TP + TN + FP + FN) Precision = TP / (TP + FP) Recall = TP / (TP + FN) F1 = 2 × Precision × Recall / (Precision + Recall) Specificity = TN / (TN + FP) FPR = FP / (FP + TN) FNR = FN / (FN + TP) MAE = mean(|actual − predicted|) RMSE = √mean((actual − predicted)²)

MetricWhat it measuresUseful whenCommon limitation
AccuracyOverall correct predictionsClasses are balancedCan hide minority-class failure
PrecisionReliability of positive predictionsFalse positives are costlyIgnores missed positives
RecallCoverage of actual positivesFalse negatives are costlyMay increase false positives
F1 scoreBalance between precision and recallBoth error types matterHides true negatives
Log lossProbability calibration qualityConfidence mattersHighly penalises confident errors
Drift scoreInput distribution changeProduction data evolvesDoes not prove performance loss
LatencyPrediction response timeReal-time systemsDoes not measure accuracy

Export and reporting

Download monitoring data, incidents, summaries, or print the complete report.

Related Calculators

Data Drift DetectorConcept Drift ChallengeAlert Threshold BuilderProduction Incident SimulatorModel Decay DefenderTraining-Serving Skew HuntLatency Monitoring GameFairness Monitoring ChallengeModel Version Tournament

Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.