Prediction Logging and Inference Data Capture Quiz

Answers and explanations appear as you take the quiz.

1. Which identifier helps link a later outcome to the exact prediction that produced it?
   • A stable request or pseudonymous entity key with model version
   • The current model's display name only
   • A random dashboard color
   • The user's plaintext password
2. Why record both event time and ingestion time?
   • To increase model accuracy automatically
   • To diagnose late, reordered or delayed records
   • To remove privacy risk
   • To replace the outcome label
3. What does data minimization recommend for sensitive request payloads?
   • Copy secrets into metrics labels
   • Store every raw input forever
   • Collect only fields needed for a defined purpose
   • Disable access controls for easier analysis
4. Why document sampling rates in prediction logs?
   • To prove labels are unbiased
   • To ensure every sampled item is positive
   • To make timestamps unnecessary
   • To understand whether the logged cases represent the full request population
5. Which signal type is best suited to aggregate request latency over time?
   • Raw model prompt text
   • Metrics
   • A model artifact binary
   • A static deployment manifest
6. What does a trace ID help connect?
   • Every user to a public identity
   • The model's training labels to its weights
   • Operations across services for one request
   • The container digest to its base image
7. Why can hashing an identifier fail to provide anonymity?
   • It changes the prediction score
   • A hash reveals model accuracy
   • Hashed values cannot be stored
   • It may remain linkable or be reidentified with auxiliary information
8. What does prediction logging not automatically create?
   • Valid, representative ground-truth labels
   • A record of captured fields
   • A link to a model version
   • An operational event history