AI advises.A person decides.

Being able to see a figure, a dashboard or a status doesn't mean anyone can act on it. Visibility still needs to be findable, understandable, and connected to something the viewer is actually allowed to do next. A screen is not a workflow.

A reporting dashboard shows the exact number a manager needs — three clicks deep in a tool nobody remembers the login for. The information exists. Nobody is using it.

A plan that's ready to execute is not the same as a plan that has executed. Readiness feels like progress because the hard thinking is finished — but nothing has actually happened in the world yet, and the gap between the two is invisible from the inside.

A policy is finished, signed off, and filed — then rolled out to a team who never see it. On paper the business is compliant. In practice nobody's behaviour has changed.

Naming something and placing it in a diagram makes it easy to talk about — which quietly starts to stand in for it existing. A named process is a plan for a process, not the process itself.

An "escalation procedure" sits on an org chart and in onboarding slides. When something genuinely needs escalating, nobody is sure who actually picks up the phone.

A system — or a person — recalling something confidently doesn't verify that it's accurate. Confidence and correctness come from different mechanisms. Only one of them can actually be checked.

An AI assistant is asked whether a client's paperwork was signed off, and answers yes without hesitation. It's working from an old summary. The paperwork was never completed.

Clearing a review means something has been judged fit to go out. It doesn't mean it has gone out. That gap is where good work quietly sits, checked twice and used by no one.

A report is approved, double-checked, and privately treated as done by everyone involved — while it sits in a drafts folder for a month, because sending it was never explicitly anyone's job.

Nobody objecting to a plan in the room is not the same as somebody having actually agreed to it. Silence gets read as consent because that's convenient, not because it's true.

A proposal goes to five people by email. Two reply "looks good." Three say nothing. The plan proceeds as if all five had signed off.

A check reporting nothing wrong is a statement about the absence of a detected problem — not a decision that something is right. Passing through unblocked is not the same as being chosen.

A piece of content clears an automated compliance scan with no flags raised. That's evidence nothing obviously wrong was found — not evidence a person decided it should go out.

A status label or a database column can describe what should happen without anything actually making it happen. The record and the reality can drift apart silently, because nothing is watching to keep them aligned.

A system has a field for "reviewed by manager," ticked as routine data entry. No review process exists behind the tick. The field looks like governance. It isn't doing any.

Fixing three parts of five and leaving two untouched doesn't produce a smaller error. It produces a document that now contradicts itself — which is harder to trust than one that was simply wrong throughout, because the reader can no longer tell which parts to believe.

A pricing page is updated for one product line but not two related ones. A reader can no longer tell whether the old prices are stale or the new one is a mistake — the correction has made the whole page less trustworthy than before it was touched.

Restricting who can see something controls its audience. It doesn't change what that something is. Material behind a login, a password or an invite-only link is still external the moment it leaves your own systems — access control is not a status change.

A working draft is shared with a small group under NDA "just to get their view." The moment it's outside the building it is an external release, whatever the access list says — and should be held to that standard, not a lower one.