SLO / error budget
Customers don't buy "good uptime" — they buy a specific promise, like "99.9% uptime, or you get a refund." That 99.9% sounds airtight, but it still allows about 43 minutes of downtime a month. Every minute you spend below the line eats into that allowance, and once it's gone the refunds and reputation damage start.
Think of the allowed downtime as a budget you get to spend each month. A little planned risk — a quick maintenance window, a tolerable blip — is fine while there's budget left. The trap is blowing the whole month's allowance on one nasty surprise, leaving you with zero room for the next thing that goes wrong. Manage the budget like the limited resource it is.
Detailed explanation
Bronze targets 99.0% (432 min/mo downtime), Silver 99.5%, Gold 99.9% (43 min/mo), Platinum 99.95% (22 min/mo). The error budget is just the inverse of the target — the slice of the month you're allowed to be down before the SLA is breached. Tighter tiers leave dramatically less room to absorb the unexpected.
Breach credits ramp 5% -> 10% -> 25% -> 50% of the monthly bill across the tiers. And there's a hard physical limit: 99.99% on a single availability zone is mathematically impossible — one shared failure domain can't deliver four nines no matter how careful you are. Selling four-nines promises forces a multi-AZ build, full stop.