• Most slos look like
    • Must be fast enough
    • Must complete (Not erroring)
    • Can maybe make this 1 slo? Slowness bakes in error or not
  • SLIs tell you what a good or bad request is
  • Want customer happiness but this is hard to measure
    • We end up measuring easy stuff
    • Fast latencies
    • Errors
    • “Variable of true or greater interest”
    • How do you measure satisfaction
    • Surveys are 1 way but usually come too late
  • Don’t want to put too much energy into the platform
    • Only want it to be reliable enough
    • … Enter error budgets
  • SLOs can be wrong. eg If you find one that by definition has been running in a degraded mode for months and customers aren’t complaining, maybe a conversation with product about whether this slo is even correct
  • Honeycomb resets SLOs when then burn through a budget completely and want to be able to keep tracking it (As opposed to it running forever as 0 budget :))