An error budget is what you can spend before breaking your SLO.
If SLO is 99.9% availability over 30 days:
- Total minutes: 30 × 24 × 60 = 43,200
- 0.1% of that = 43.2 minutes of downtime allowed
- That's your error budget
When budget is healthy, you can take risks: deploy more, run experiments, accept some instability for feature velocity. When budget is depleted, you stop shipping features and focus on reliability.
This is the bridge between SRE and product. Product wants velocity. SRE wants reliability. Error budgets make the tradeoff explicit and data-driven.