Main idea

A service commitment that excludes outages caused by a resource under the customer's control — its SQL Server database, its network, its keys — looks like it protects the vendor. It only protects if you can objectively establish that an incident really did come from there.

Without that capacity for attribution, the clause is just text: at the moment of the outage, each side maintains the cause sits with the other, and it is the balance of power that decides, not the contract.

A split of responsibilities written with no means of measurement is therefore not a split. It is an intention.

Why it matters

This moves what has to be negotiated. The usual discussion is about the scope of the exclusions; what decides their worth lies elsewhere — in the signals available to investigate an incident.

The consequence is an order of operations: instrument first, draft second. A clause written before the measurement exists creates imaginary safety.

Nuances and limits

Attribution is never total: some outages have entangled causes and stay undecidable, whatever the instrumentation.

And the measurement itself can be contested if it comes from only one of the two parties.

Open questions

• Do attribution signals have to be shared, or even produced by a third party, to be enforceable?