Day 32 10 min
Measuring a service
1 / 6 · ≈ 15 min left
The problem
Mia is on call for an online shop's checkout API. Last week her phone woke her three times for "CPU above 80%". Each time, customers were paying normally and she went back to bed.
On Thursday, the payment provider changed a TLS certificate and 4% of checkouts failed for two hours. No alert fired, because the CPU was fine. Support heard it first, from customers.
The team wants pages that fire when customers cannot pay, and only then. That means choosing what to measure, what number counts as good enough, and when a human must wake up.