The reason to restate a rule that old is a product that new. On August 31, AWS opened a public preview of its managed multicloud interconnect with Microsoft Azure — Azure’s side ships as Azure Multicloud Interconnect — completing a set that already includes Google Cloud and Oracle, both generally available since spring and summer. A private, dedicated-bandwidth link between two clouds is now provisioned the way a storage bucket is: from a console, in an afternoon, by whoever holds the permissions. AWS has offered a 500-megabit tier of it at no charge since May. Until this year, the same connection meant a carrier-sold circuit, a colocation cross-connect, and somebody’s signature on an agreement that priced all of it.

Look for the Exhibit behind the new link and the search comes up short. The link is real, the charges above the no-charge tier are real — and the pricing lives on a published rate card the provider maintains and can revise, folded by reference into terms your team accepted with a checkbox. Nothing in that arrangement is improper. It is simply not the arrangement the discipline was built for. The rule says a charge is disputable when it cannot be traced to an Exhibit. Here the charge is traceable — to a document the vendor alone controls.

Where the invoice lands makes the gap wider. Carrier-sold cloud connections have always billed onto the carrier master account, then reached the cloud team as an internal chargeback from accounting: the team that reviews the carrier invoice did not place the order, and the team that placed the order sees a chargeback line, not the charge. Provider-managed interconnects reverse it: the charge lands on the cloud bill, inside the cloud team’s console, and the telecom team — the one that keeps the rate cards, the Exhibits, and the disconnect discipline — never sees it at all. Either way, the desk that could test the charge and the desk that receives it are not the same desk.

Then there is the tier at no charge. In a carrier agreement, a no-charge period is written down in advance: it has an end date, and the price that follows is on the page. Estates miss the date anyway: the invoice has read zero for months, and the first month it does not looks like a small new line item rather than the end of a term. A preview is the same thing with less paper. Preview pricing ends when the provider says it has, and a link provisioned in an afternoon in September is a production dependency by January. Whoever budgets for it needs the date, and the date is not on your calendar unless someone put it there.

And the circuit the new link replaces does what carrier circuits have always done: it bills until a disconnect order is placed and worked. Not until the traffic moves. Not until the project closes. Until the order.

So the discipline has to be rebuilt for connectivity that no Exhibit governs, and it rebuilds on the three things the buyer still controls. The order: what was provisioned — bandwidth tier, regions, the date — captured the day it was placed, because the console will not remember it for you. The rate card as it stood on the invoice date, saved, so that a revision shows up as a variance instead of as the new normal. And the inventory: every link tied to a live workload, every displaced circuit tied to a worked disconnect order. Held against those three, a cloud-connectivity charge is as testable as any line on a carrier invoice. Across 31 months on a multi-brand retail portfolio, that test — the charge against the document that authorizes it — produced $1,852,314 in credits from the carriers, at an 87% collection rate. BearGuard, our managed governance process, applies the same test every cycle, which is why we can say which document each charge answers to before anyone asks.

The Exhibit did not disappear when the connection became a click. It moved — into a rate card the vendor maintains, an order your team placed, and an inventory only you can keep. The terms are still somewhere. Governance is the work of knowing where.