The Floor the Score Cannot See

The staffing optimizer filled the weekend in four minutes. Green coverage. No open shifts. Ken, who still walks the floor on Thursday nights, knew two of the names wouldn't hold: one had already told him, off the roster, that she was done with doubles. The other had been a rumor the model had never been told to treat as a constraint. He had twelve minutes before the schedule locked. He signed. The weekend broke by Saturday noon.

The business case counted minutes saved. It didn't count the knowledge that never entered the model.

The Human Overhead of Agentic Systems is the second org chart: reviewers, exception-handlers, the conscientious person who absorbs what the vendor slide forgot. This is the quieter inventory inside that chart. Not hours of review. Facts that live in a person. Who already said no. Which client won't wait. Which name on the roster is a courtesy and not a body. The optimizer can't hold those. It wasn't given them. Fluency will staff the gap unless a person is still allowed to be a gate.

Labels leave the floor out

Scores are loyal to their labels. If "available" means a box ticked in the last cycle, the model will keep ticked boxes. It won't know the conversation in the parking lot. It won't know the no that was never written down because writing it down felt like disloyalty. That's a claim about where knowledge lives in a team that still has a floor.

When the Shortlist Is Fluent and Still Unfair is about names the ranking never saw. This is about names the ranking did see, and treated as real. Both are a failure of what gets to count. Fairness work asks who is missing. Street knowledge asks what the people who are present already know that the file doesn't.

Leaders treat that knowledge as folklore. Useful, unofficial, slightly embarrassing next to a dashboard. Then they're surprised when a green roster fails. The surprise is the tell. They had already decided that only labeled facts were allowed to change the call.

Trust Spent on a Recommendation You Cannot Explain is the twelve-minute signature on a scored shortlist. Ken's signature is cousin to that: he couldn't say, in the language the lock-screen would accept, why two names were fiction. So he didn't. Explainability as a leadership skill is four sentences a person will sign. Floor knowledge has to be allowed into sentence two (on what basis) or the other three are theater.

Street knowledge belongs in the file

Practical intelligence, in the old language, is the skill of seeing which room the fluent answer actually belongs to. The optimizer belongs to coverage. The weekend belongs to bodies who will or won't show. Those are different rooms. A person who still walks the floor is holding the constraint the system was never built to hold. Calling that resistance misses the job.

If you punish that person for slowing a green lock, you will get a faster lock and a quieter floor. The knowledge won't disappear. It will stop arriving in time. Then you will call the break an exception, and you will staff the exception with the same person, unpaid, as human overhead already warned.

You don't need a new model. You need a line in the file that the lock can't skip: three facts a named person on the floor knows that the score can't. If any of them would change the call, they're the call. If none of them would, you have tested the score against the building, which is the only audit that building will respect.

This week

Pick one scored decision that will ship: staffing, routing, a ranked list or a spend. Sit with the person who still has the floor. Write three facts the model can't know. Put them in the file before the signature, not in the incident report after. If there's no field for them, that's the finding. Make a field. Then decide whether the score still holds.

The person remains the one who can be asked why. A roster that can't see the floor has already answered, and it has answered with a fiction that looks like coverage.

The five questions in the Ethical AI Leadership Decision Toolkit are built for this kind of page. Field notes continue in EI Leadership Insights.