The Downtime Report That Counted the Same Minute Twice
### P2 · The reliability report does not reconcile Checkout's one-hour report claims 1,420 unavailable seconds, but overlapping alerts cover only 940. A retry and a nested component alert are being counted as extra outages.
Your task: Fix only covered_seconds() in outage.py. Clip half-open intervals to the report window and measure their union. Preserve gaps, handle duplicates and nesting, and leave inputs unchanged. Work on endpoints, not every second.
Investigate: Run python monitor.py, then python -m pytest -v. The starter is intentionally wrong; the monitor should reconcile after your repair. Open a PR against main with your reasoning.
Learn: interval union, sorting, boundary semantics and regression testing. Python 3.12, standard-library runtime, pinned pytest test tools.
The incident and timeline are authored fixtures. This verifies elapsed coverage aggregation, not real production availability or contractual SLA compliance.
Skills you'll prove
Interview Tip
I treated alerts as intervals, not independent durations. I clipped them to one half-open reporting window, ordered a copy, and accumulated covered segments without double counting. Nested intervals could not shrink the active end, gaps stayed uncovered, and tests covered duplicate alerts, large timestamps and unchanged inputs.