Build a percentile table

Constraints

1. 1Use the nearest-rank definition: the pXX value is the delivery at position ceil(n * XX / 100) once all deliveries are sorted by time. It always returns a time some delivery actually took.
2. 2Do not interpolate between neighbouring rows. An interpolated p90 is a number no delivery ever recorded, and it cannot be defended in an SLA review.
3. 3This engine has no PERCENTILE_DISC. Build the rank arithmetic yourself, and note there is no CEIL for integers either — integer division truncates, so add 99 before dividing by 100 to round up.
4. 4Four rows out, one per percentile, ordered by label.

Worked example

There are 26 deliveries. For p90 the target rank is ceil(26 × 0.9) = ceil(23.4) = 24, so p90 is the 24th fastest delivery — 96 minutes. In integer arithmetic, (26 * 90 + 99) / 100 = 2439 / 100 = 24, because the division truncates. Adding 99 first is how you round up without a CEIL function. The punchline is the comparison: the mean delivery time is 48.1 minutes, which lands between p75 (45) and p90 (96). The average delivery is slower than three quarters of all deliveries.

What this tests

Whether you know what a percentile actually is, rather than which function name computes one — and whether you can express a ceiling in integer arithmetic.