CSDM crawl, walk, run — what each stage actually requires

The CSDM adoption stages are usually read as a checklist of entities to populate: crawl gets you these classes, walk adds these, run adds these. Read that way they produce estates that have the records and none of the capability.

They work better read as what you must be able to defend before the next stage means anything.

Crawl — can you trust that a record is one thing?

The crawl question is identity. Are the CIs distinct, do they resolve consistently, and does the same device come back as the same record every run?

You have left crawl when duplicates have stopped regrowing — not when they have been cleaned up once. If you can clean and they return, you are still in crawl, whatever the entity checklist says.

Walk — can you say where a thing sits and why?

Walk is where judgment enters, and it is where most programmes quietly stall. Applications get mapped to services, owners get assigned, scope gets drawn.

None of that is discovered. It is decided, usually in a workshop, often once. The thing that determines whether walk holds is not how many relationships you asserted but whether anyone can say, a year later, who decided and on what basis. If they cannot, those relationships will be re-litigated on the first big incident and the programme will go backwards.

Run — does anything break when the model is wrong?

Run is when the model carries consequence: change assessments reference it, impact analysis depends on it, automation acts on it.

That is the real threshold, and it is dangerous to cross early. A model nobody relies on can be wrong for years harmlessly. The moment something automated acts on it, every error has a cost — so the honest precondition for run is not completeness, it is that you can tell which parts are reliable and which are not.

The failure that is worth naming

Emitting run-stage artefacts into a crawl-stage programme creates noise, not maturity.

A full service model handed to a team that cannot yet keep identity stable will be wrong within a quarter, and the wrongness discredits the whole exercise. Stage the output to match where the programme actually is, not where the slide says it should be.

How to tell which stage you are in

Ignore the entity checklist and ask one question: what currently breaks when the model is wrong?

If the answer is “nothing, really”, you are earlier than you think — and that is useful information, not a failure.

The whole CSDM 5.0 model with what arrives at each stage is in the adoption stages documentation.