When reading the Nines are Not Enough paper, I had some ridiculous ideas. The paper says that specifying SLOs is hard. I wonder if it is possible to use a gaming process to help develop more reasonable SLOs. For example, it might be financially incentivized to be a “professional SLO violation hunter”. The main task of that job would be to find ways to make cloud providers unable to meet their SLO and sue them for money. And cloud providers would then try to find a better SLO so that they wouldn’t be sued again. And through this iterative gaming process, I suppose some brilliant ideas would show up?

My guess is this would probably work for performance SLOs, like IOPS, packet loss rate, etc. But probably it wouldn’t work for “Compositional SLOs” like multi-datacenter fault tolerance. Also, for those SLO violation hunters to find out the violations, the SLIs need to be observable from the customer side, which, as the paper mentions, is not always achievable without the support of the provider.