A percentile is the value below which a given share of the data falls: the 90th percentile of delivery times is the time 90 percent of orders beat. pandas .quantile() takes a fraction (0.9); np.percentile() takes a percent (90).
A z-score says how many standard deviations a value sits from the mean: (x - mean) / sd. A z of 2 is unusual in roughly normal data.
To flag outliers in skewed data, the IQR rule is safer than z-scores, because the outliers themselves inflate the SD. Values above Q3 + 1.5 × IQR (or below Q1 - 1.5 × IQR) are flagged.
A share of values below a cutoff is just the mean of a True/False column: (s < 45).mean().