In an internal cockpit, the operator reviews cards. Each card carries a decision: follow up with this contact, send this proposal, mark this file as handled. A swipe to the right means: yes, do it.
For several weeks, that gesture did nothing. It changed a status row in the database. The promised action never happened.
An invisible and expensive gap
On the surface, everything worked. The card turned green, the list moved forward, the dashboard looked alive. The problem was lower down: the follow-up had not been sent, and nothing said so.
This kind of gap is dangerous because it does not look like a failure. A failure blocks. Here, the system keeps moving. It lies quietly.
The cost is not in the code. It is in the trust granted to a tool that says it has acted.
The rule
One card equals one action equals one gesture. Approving a card does not change a status. Approving a card executes the thing.
If the action cannot run, the card must remain open. If the channel is unavailable, the card must say so. If a human approval is missing, it must stay visible. The system is not allowed to say done until it is done.
Why the rule holds
A cockpit is not valuable because it displays many buttons. It is valuable because there is a strict correspondence between a human intention and a real effect in the system.
The rule sounds narrow. It prevents a common drift: building decision interfaces that are only decorative screens on top of disconnected scripts.
A gesture that does nothing is a lie. A green status without a downstream effect is one too.