A large organisation wants consistent architectural decision practice across many teams. What should be standardised and what should not?

What should be standardised

1. That significant decisions are recorded. Not the format — the practice. A team that records decisions in its own style is far better than one that records none.

2. The minimum content: the decision, the options rejected and why, the constraints that forced it, and the revisit condition. These are what a future reader needs, and their absence is what makes records useless.

3. What counts as significant, published in advance — the same trigger set that governs review: a new data store, a new external interface, regulated data crossing a boundary, a new failure domain, a dependency added to a critical path, an irreversible choice.

4. Discoverability. Records live with the code, linked from the components they govern, so the reasoning is encountered by the person changing the thing.

5. That disagreement is recorded, so a decision made over objection is visible as such.

What should not be standardised

The format, the template, the tooling, and the depth. A mandated template with fifteen sections produces compliance documents nobody reads. Teams that own their practice maintain it; teams complying with a standard produce artefacts for the standard.

Also not: the decisions themselves. Central standardisation of technology choices across teams with different problems produces either exceptions or bad fits — the appropriate central control is a paved road that makes the standard choice easier, plus guardrails for the small set of genuinely severe cases.

The practice that makes it stick

Records written before deciding, circulated for comment. The value of writing is largely in the thinking it forces, and a document produced afterwards captures a conclusion without the reasoning that would have improved it.

The measure

How often decisions are re-opened without new information. A decision revisited because a trigger fired is the system working; one revisited because nobody remembers why it was made is the failure the practice exists to prevent — and it is the dominant hidden cost in most large organisations.