An error budget is the acceptable amount of unreliability inside an SLO — 1 minus the objective. A 99.9% SLO means 0.1% of requests can fail without breaching. That 0.1% is the budget the team can spend on deploys, experiments, and controlled risk.
Error budgets convert reliability from a wish into a currency. Each unavailability event, latency breach, or deploy-related regression spends budget. When the budget is healthy, teams can ship aggressively. When it is depleted, deploys freeze and every engineer focuses on reliability work until the budget recovers.
The standard implementation uses a rolling 28- or 30-day window with burn-rate alerts: a fast burn (10% of monthly budget in an hour) pages the on-call; a slow burn (25% over a day) opens a ticket.
Error budgets end the eternal fight between product and SRE. Instead of 'stop shipping so much' vs 'stop being paranoid', the team follows the number. Fast, sustainable, and shared.