Skip to main content

Burn windows and reliability evidence

This required 20–30-hour package supplies guided Lab 1's executable alert core and strengthens Lab 4: lessons/predictions 5–7 hours, evaluator and telemetry experiments 9–15, assessment/repair 6–8. Reuse original service and monitoring work. Estimates await semester reconciliation and learner pilots.

Prerequisite bridge​

Review ratios, weighted aggregation, request eligibility and evidence coverage. Revisit Module 01 admission/backlog and Semester 07 missing-observation behavior. Keep availability, latency, resource saturation and collection health distinct.

Sequence​

  1. Weighted burn.
  2. Window coverage.
  3. Action and containment.
  4. Implement burn-window-contracts.zip in Python 3 standard library. Run python3 -B check.py starter before reference calibration.
  5. Complete RE1–RE6.

The reference tests 625 four-bucket cases, incomplete history, minimum traffic, inclusive threshold and invalid inputs. It is one paired-window evaluator, not a complete alerting/monitoring platform. Original actual SLI data, fast/slow policy, saturation and overload experiments, game day and incident follow-up remain required.