Build1 publisher2 min readPublished
An SRE roadmap hangs its case on a 47% low-performer share from 2018
The post blames engineering discipline for the persistence of DORA's Low cohort and prescribes SRE. Its Phase 0 exit gate is four golden signals in Prometheus for three production services, sourced from Envoy.
The Engineer · Build desk
What happened
- By 2023, the post says, the Elite and High shares had grown while the Low cohort had not emptied, leaving a wider gap between organisations that adopted modern delivery practices and those that did not.
- The author reads that persistence as an engineering discipline problem, not a tooling or process problem, and names Site Reliability Engineering as the discipline that closes it.
- In the post's reading of the Four Key Metrics, MTTR counts the time to restore a failed deployment, and the reliability of the running service sits outside its scope.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint A team cannot use its delivery scorecard as a reliability KPI, because the four metrics stop at the deploy and the five reliability questions the post lists need a separate instrument.
- cost Sourcing the golden signals from the proxy puts mesh coverage on the critical path, and the bill lands on whoever runs the platform.
- decision The ordering decides budget sequence: observability work has to be funded and trusted before an error budget policy can be pitched, so teams leading with governance are arguing against the roadmap's own gate.
- contradiction The causal claim rests on a comparison the post reports qualitatively, so the cohort statistic on its own cannot separate stalled adoption from a change in how organisations were scored.
Deployment frequency, lead time, change failure rate and MTTR all describe one object: the path from commit to production. The post is explicit that they measure how quickly and safely an organisation can make changes [5]. Alongside them it lists five questions about reliability: what "reliable" means for a specific service from a user perspective, how much unreliability is acceptable and who decides, and the investment ceiling for reliability. Two more follow: how to detect degradation before it reaches users, and how to balance release velocity against production stability [7].
The Phase 0 gate is written as an observable condition, which is the part of this roadmap I would keep. Latency, traffic, errors and saturation have to exist in Prometheus for at least three production services, and the metrics have to come from the Istio Envoy proxy, not from code the service team writes [9]. Instrumenting three services is the cheap half. As written, a service with no Envoy proxy in front of it cannot satisfy the gate at all [14]. Before adopting this sequence, count how many of your critical services already sit behind a proxy.
The statistic underneath the thesis is a single line. About 47% of the organisations in the DORA research study were classified as Low performers in 2018, the year Forsgren, Humble and Kim published Accelerate [1]. By 2023 that number "had not materially decreased", the post says [2]. It does not give the 2023 percentage or identify which DORA report the 47% came from. For the comparison to carry the claim that transformation stalls on discipline instead of tooling [4], the 2018 and 2023 populations would have to be comparable samples scored against comparable cluster boundaries. Subtract the Low band from the whole and roughly 53% of the 2018 population sat above it [13], with the growth the post reports concentrated at the Elite and High end [3].
I think the weaker half of the argument is the statistic and the stronger half is the list of gaps. SLIs and SLOs, error budgets, observability engineering, toil elimination and the error budget as a change gate are all checkable on one service [8]. Either an SLO exists, or it does not; either a breach stops releases, or releases continue. The post puts hard ordering on that work. No SRE governance before Phase 0 is complete, and no SLO enforced before the measurement is trusted. No error budget policy proposed, either, before stakeholders understand what an error budget is [10]. DevOps comes first, and SRE builds on it [11].
What to watch
- A later DORA report giving a Low-performer share scored on the same cluster boundaries as the 2018 figure would settle the trend question.
- Whether teams publishing Phase 0 evidence can source golden signals from Envoy for services that never joined a mesh.
- Any reported case of an error budget breach actually blocking a release, which is the test of the change-gate claim.