DORA Methodology

Alexander Polomodov and Igor Kurochkin examine the method behind DORA’s annual reports. Igor has followed the research since its early years and joined an open terminology review. State of DevOps and Accelerate led to the evolving capability → performance → outcomes model.

Research starts with a goal, model, and hypotheses; questions come later. Terms such as small batch need consistent meaning. Since 2023 one large model has given way to smaller capability models. The survey must remain short and unambiguous, while Likert scales make answers comparable.

Participants arrive through an open link, partners, and invitations, so self-selection and snowball bias are unavoidable. DORA documents these limits and randomly routes people through survey branches. Constructs use multiple questions: claimed trunk-based development conflicts with answers that reveal long-lived branches.

Westrum’s culture model shows how an abstract property emerges from observable traits. Statistical models separate training from validation and move from correlation toward causal hypotheses. Bayesian reasoning addresses confounders and aggregation from people to organizations, but without raw data some of the method remains hidden know-how.