Support and managed-services teams get asked to take over platforms that were launched with enthusiasm and documented with none. Before accepting one — or before your quarterly review of one you already run — a structured audit pays for itself many times over. What follows is a twelve-check PlatformOps audit built on the definitions in the PlatformOps reference published by xOps School, arranged so that each check names the evidence to demand, not just the question to ask. The distinction matters: platforms are audited to death with intentions, and intentions are exactly what an audit must not accept.

A note on posture before the checklist. An audit of an internal platform differs from a security review or a cost review: those ask whether the platform is safe or cheap, while this one asks whether it is actually operated — whether promises exist, whether somebody carries them, and whether the platform’s history shows commitments being kept and retired on schedule. In practice that question decides more handover outcomes than any vulnerability scan, because unoperated platforms fail slowly, politely, and then all at once.

Before the PlatformOps Audit: Scope and Ground Rules

Three ground rules keep the exercise honest. First, audit artefacts, not atmospheres — a check passes when a document, a log or a number is produced, never when someone explains confidently that it exists. Second, sample rather than accept: pick two or three capabilities at random and trace them fully rather than skimming all of them shallowly; the deep trace is where the discrepancies live. Third, separate the build from the run. A platform may have been constructed beautifully and still fail this audit, because PlatformOps is the operating discipline and operations are what fail silently in managed handovers.

On timing: run the full audit at every handover between providers, once a quarter for platforms under active development, and after any incident that crossed a boundary — platform fault treated as application fault or the reverse. An audit is also the correct first week’s activity when you inherit a platform cold, ahead of any improvement work, because remediation without a baseline is just opinion with a budget. Twelve checks take roughly a day to run properly, two if the documentation resists being found, which is itself a finding worth recording.

Running the PlatformOps Audit: Checks One to Four

The opening block covers foundations — the paperwork that everything else depends on.

1. Charter and scope. Demand the one-page charter. Verify it states what the platform covers, what it explicitly refuses, and how teams escalate disagreements. Evidence: the document itself, dated, with a decision log showing it has been consulted — not merely written.
2. Catalogue ownership. Pull three catalogue entries at random and confirm the named owner knows they are listed. Then cross-check: compare declared consumers for one capability against actual access logs. Research cases exist where a capability declared five consumer teams and logs showed seven real callers — the kind of gap that turns into an outage with an audience.
3. Lifecycle register. Every capability should carry a public stage — proposal, trial, general availability, deprecation, retirement. Evidence: the register, plus one example where a stage change triggered an actual communication to consumers.
4. Service levels. Each capability needs a published number and a review cadence. Evidence: the levels, plus the minutes or dashboards of the reviews where they were checked against reality. A level of service that has never been measured is a wish with formatting.

Checks Five to Eight: Promises and Arithmetic

The second block tests whether the platform’s commitments are real, countable and bounded.

5. Error budget arithmetic. Where a level like 98 per cent success exists, verify someone has computed its consequences: across 2,400 requests, that promise permits 48 failures. The arithmetic is trivial and the discipline is not — teams that compute budgets spend them deliberately; teams that do not, spend them by accident.
6. Support tiers. Confirm each capability documents who responds, on what timeline, and through which channel. Evidence: two resolved tickets traced end to end, matching the tier they were supposedly handled under. Watch for the quiet merger of tiers — documents promising a premium response while the rota shows one engineer covering both tiers is a partial fail at best.
7. On-call boundary. Verify in writing that application teams carry their own services and the platform team carries platform components. Then check the paging history: any platform engineer paged for an application defect is a boundary violation, and a pattern of them means the boundary exists only in slides.
8. Golden path reality. The path — the supported route for the most common workload — should exist, be documented, and be measurably used. Evidence: coverage figures by team, and the exit records of anyone who left it, with reasons. A path with no recorded departures is either perfect or compulsory; both deserve scrutiny.

Checks Nine to Twelve: Boundaries and Evidence of Life

The final block examines how the platform behaves toward its consumers and what history says about it.

1. Consumer testing against objectives. Ask whether the platform team ever verifies its promises by subjecting consumers to controlled failure. Google’s engineers famously ran synthesised outages of the Chubby lock service — deliberately, with notice — to confirm which services truly depended on it and which merely claimed to. A platform that has never rehearsed its own failure is selling untested assurance.
2. Staffed deprecation. Find the most recently retired capability. Evidence: a dated notice, a migration plan, and a named person who helped users across. Warnings alone do not migrate anyone; if retirement depended on a changelog entry, mark the check failed and expect a long tail of zombies in the estate.
3. Operational metrics, reviewed. Demand the numbers and their review history: adoption, self-service rate, time to first delivery for newcomers, satisfaction, and delivery outcomes. DORA’s research supplies the analytical frame — platform programmes can depress throughput and stability when run badly, so the audit must see delivery metrics trending, not just usage counts rising.
4. Voluntary adoption. Finally, ask how the last three consuming teams arrived. Mandated users produce compliance, which mimics adoption on dashboards while hiding walkaways. Evidence: interviews with two team leads, separately, on why they joined and whether they considered not doing so. The hesitation in their answers is data.

Scoring, and What Failure Looks Like

Score each check pass, partial or fail, and resist the managed-services instinct to average them. Platforms fail asymmetrically: a missing on-call boundary or an unstaffed deprecation function is not one weak point among twelve, it is the fault line the next incident will open. A clean audit shows twelve passes with artefacts; a realistic one shows nine or ten with a remediation list, owners, and dates. What should end the conversation outright is a pattern — missing charter, missing register, missing levels — because that trio describes a platform that was built and never operated, and inheriting it means accepting the debt.

Write the report for two readers at once. The provider needs the remediation list, ordered by risk rather than by check number. The client needs one page: what was verified, what was assumed, and what the platform cannot currently promise. That second page ages well — when a dispute arrives in six months, it is the document both sides reread first, and the presence of an honest baseline changes the tone of every conversation that follows.

For teams new to this kind of scrutiny, the underlying principles overlap heavily with the delivery fundamentals described in these DevOps principles for effective software delivery — ownership, measurement and small verifiable steps, applied to the platform layer. The audit also maps neatly onto role boundaries: much of it is effectively the platform counterpart of the responsibilities catalogued in this guide to DevOps role titles and job roles, which helps when a handover is ambiguous about who was supposed to hold which promise.

One closing note for individuals rather than organisations. Engineers who can run this audit — calmly, with evidence, without theatre — are rare, and the skill is portable: it is essentially due diligence for internal systems. Documenting a platform audit you have led, findings and remediations included, is exactly the kind of artefact that strengthens a high-impact DevOps engineer’s portfolio, because it demonstrates judgement rather than tool familiarity. Audits, like platforms, are judged by what they leave behind.