80. An answer was correct when an evaluation was created but is wrong after a policy changes. How would you maintain the benchmark?
MediumMachine learning basics
Technical
Show model answer
Record the reference date, authoritative source and intended time scope for each time-sensitive task. Decide whether the task asks about the historical situation or requires the latest valid information.
Update or retire affected references through a reviewed process while preserving benchmark versions for comparison. Separate a stale reference from a model error, and avoid silently changing labels midway through a candidate comparison.
What interviewers listen for
- Version dates and authoritative references
- Distinguish historical from current tasks
- Review label changes between comparisons