An explicit agreement between engineering, product, and business on what 'acceptable' performance looks like per page or feature (e.g. all pages under 300ms, some pages under 100ms, others up to 1s). Panelists argue defining SLAs before optimization avoids wasted energy; without them, anything can be 'fine'. Stephen notes renegotiation is reasonable as context changes. Caio emphasizes SLAs should be defined across teams and reflect user-perceived value (optimizing an async background update the user can't see is low-value). Maciej counters that he rarely sees formal budgets 'in the wild' and recommends craftsmanship-driven ad-hoc time-boxed optimization when no SLA exists.