This case study concerns a fictional team — a composite assembled from patterns seen across many real organisations, with details invented for clarity. Call them Fietspad, a twenty-two-engineer product group at a Utrecht logistics-software firm. Their year with the delivery maturity audit is told here in full, including the quarter it nearly collapsed, because the failure contains most of the lessons.
January: the uncomfortable baseline
Fietspad’s story began, as these stories usually do, with a bad release. A pricing defect reached customers on a Friday, the rollback took eleven hours, and the retrospective produced the sentence that changed the year: “We keep saying we’re pretty good. Does anyone have evidence?”
Nobody did. So the engineering manager, Sanne, proposed an experiment: a formal maturity audit of the team’s delivery practice, scored and repeated every quarter. She had encountered the pattern while researching governance tooling — SCMGalaxy, a training and consulting company in the delivery space, markets SCMGalaxy OS around structured assessment: a hundred-item inventory across defined practice areas, scores constructed so they do not depend on who holds the pen, and improvement plans phased over the months that follow. Fietspad could not justify an enterprise platform for one team, but the architecture was free to copy. They wrote thirty evidence-based questions across five areas, fixed the marking scheme in writing, and scheduled a full day.
The baseline landed hard. Release practice scored lowest: deployments required four manual steps, two of which lived in a senior engineer’s personal notes. Security hygiene came second-worst — dependency updates happened “when someone remembered,” which the evidence showed meant rarely. Two areas scored respectably; nobody cared. The low numbers consumed all attention, which, Sanne noted later, was exactly their function.
February to March: the first roadmap
The team converted the two weakest areas into a staged plan: a 30-day list, a 90-day list, and a 180-day list. Thirty days: script the manual release steps, retire the personal-notes dependency, enable automated dependency alerts. Ninety days: a rehearsed rollback procedure with a measured time target, and review rules that no change merges unexamined. One hundred eighty: a security-practice training block and a documentation culture that would survive personnel changes.
Two design choices proved decisive in hindsight. Every line had a single named owner — not “the platform group” but “Joris.” And the re-score date went into the calendar before any work began, which converted the roadmap from aspiration into appointment.
April: the second score, and a surprise
The April audit used the same thirty questions, same marking, same panel. Release practice had climbed sharply — scripted deploys, one rollback executed in nineteen minutes. Security had barely moved. The surprise was the reaction: no argument. In previous improvement pushes, progress had been debated endlessly, every claim met with a counter-anecdote. Two comparable numbers ended that genre of meeting. The security stall was traced within an hour: the alerts existed, but nobody owned triage, so they accumulated unread. One sentence in the next roadmap fixed what three months of good intentions had not.
Precision aside, something cultural had started. Junior engineers, who had experienced earlier initiatives as weather from above, began citing scores in planning discussions. One, Femke, rebuilt her personal development plan around the team’s weak areas — a shrewd reading of how engineers progress from junior to senior: visible competence in the team’s measured gap is the fastest route to being noticed.
July: the quarter it nearly died
Honesty requires the bad quarter. July’s audit collided with a product deadline, and the panel — tired, rushed — marked generously. Scores rose across the board. For about two weeks this felt pleasant. Then a staging incident contradicted the fresh observability score so flatly that the team lost faith in the entire instrument. “If the number lies once,” as one engineer put it, “why keep the number?”
Sanne’s response saved the practice. She declared the July round void, re-ran it with an outside engineer added to the panel, and wrote a new rule into the audit charter: a flattering score is a defect, and any panellist may call for re-marking. The re-run restored the honest, lower numbers — and, oddly, restored morale with them. The team had learned the instrument’s real fragility: not mathematics, but courage. Deterministic marking only works when the people applying it refuse to be kind.
October: compounding begins
By the fourth round, the audit had stopped being an event. Preparation took hours instead of days because evidence was now organised continuously — pipelines logged what the questions asked, documentation lived where the panel would look. Three of five areas scored strongly; the roadmap thinned from twelve items to five. The October retrospective recorded effects nobody had planned for:
- Onboarding time for new hires had dropped — the scorecards and roadmaps formed an honest, current map of the operation that no wiki had ever managed.
- Training spend had become targeted. Courses were chosen against measured weaknesses, in the spirit of the capabilities engineers genuinely need this year, rather than by catalogue browsing. Two engineers pursued security certifications aligned to the stubborn weak area — a move consistent with the career case for certification, but chosen for the team’s gap first.
- Leadership conversations changed register. Budget requests arrived with before-and-after scores attached, and were approved faster.
December: the second year begins differently
The year closed with a planning exercise that would have been unthinkable twelve months earlier. Sanne brought four quarterly scorecards to the engineering review and let them structure the conversation. Budget asks were ranked by which weak area they attacked; two tooling proposals died on the spot because no score justified them; the security training line survived precisely because its area’s stubborn flatness was visible across three rounds. The team also made its first deliberate change to the instrument itself — retiring two questions that had maxed out for three consecutive rounds and drafting replacements for the newly exposed next layer of weakness. They versioned the questionnaire, recorded why each change was made, and accepted that the affected areas’ histories would reset. Instrument evolution, they had learned, is legitimate exactly once a year, performed in the open, and never in the week before a scoring session.
Femke, the junior who had rebuilt her development plan around the gaps, presented the December summary herself — her first leadership-room appearance. The slide everyone photographed held nothing but ten numbers: five areas, two points in time. It was, Sanne remarked afterwards, the cheapest credibility the team had ever purchased.
What the composite teaches
Fietspad is fiction; the mechanics are not. Four findings deserve the summary treatment.
The instrument must be boring. Fixed questions, written marking rules, unchanged between rounds. Every interesting modification destroys comparability, and comparability is the product.
The date does the work. A scheduled re-score altered behaviour all quarter. The audit’s power is mostly anticipation.
Honesty is the failure mode to engineer against. The near-collapse came not from bad process but from kind marking. Build in an outside voice and a void-and-rerun rule from the start.
The score is a means. What Fietspad actually acquired in a year was a habit: measure, stage the fixes with dates and owners, execute, measure again. The numbers merely made the habit undeniable — to the team, and to everyone funding it.