Skip to content

The PDSA Loop

PDSA — Plan, Do, Study, Act — is W. Edwards Deming’s continuous-improvement loop. pdsa implements it as a working feedback cycle and records each round to a graph so the learning accumulates.

StepQuestionWhat pdsa records
PlanWhat will we try, and how will we know it worked?Your intent + a verifiable expected evaluation
DoWhat did we actually do?The execution, organized against the plan
StudyWhat did we learn?A verdict (met/partial/unmet) + the measured actual
ActWhat do we change next?The next improvement action; a REINFORCES link if needed

The most important distinction: the third step is Study, not Check.

  • Check asks a yes/no question — “did it pass?” — and stops there.
  • Study asks “what did we learn?” — it compares the result to the hypothesis set in Plan and extracts knowledge, whether the outcome was met or not.

pdsa enforces this by making Plan produce a verifiable expected evaluation, and making Study judge against it and narrate the learning. A partial or unmet verdict isn’t a failure of the tool — it’s information.

This isn’t just bookkeeping. The reference implementation in AkkaGraphLoop.Core models the loop as an actual Akka.Streams feedback cycle (PdsaLoop), the trickiest kind of stream graph — a cycle that must stay live (no deadlock) while it feeds its own output back as the next input. The learning project this tool grew out of exists to study exactly those cyclic graphs.

PDSA descends from Walter Shewhart’s plan-do-see cycle, which Deming reshaped and carried to post-war Japan, where it underpinned the quality movement (Toyota and others). Deming later insisted on Study over Check precisely because improvement comes from learning, not inspection. See the project’s essay PDSA — History, Theory, and the Quality Legacy.