The scenario
Evict samples outside the window and rank percentiles correctly so a stale burst stops skewing the dashboard.
A failure to reason about
A dashboard still includes an old slow request after it should have left the observation window.
Illustrative context for the topic. The exercise’s full requirements arrive in the workspace.
Questions to bring to your review
- Which caller makes the reported behavior visible?
- What focused regression would distinguish the current behavior from the requirement?
- What existing behavior must remain intact after the change?
Skills you will practice
- Percentiles
- Sliding windows
- Data aggregation
Read the existing code, reproduce the reported behavior, review proposed changes, and verify your final implementation. You can use AI assistance while staying responsible for the result.
Prepare with a worked example
- How to review AI-generated code
A worked review of a cache patch: establish the contract, find a counterexample, verify the fix, and explain the evidence.
See how nrml evaluates an attempt, including the limits of automated evidence.