The side no arithmetic can move

Three classes, and what they have in common

Four of Serdana's nine rows failed the inequality. Three of them failed for reasons that would survive any improvement in the technology, any fall in p, and any rise in D. They are not close calls. They are actions whose arithmetic cannot be completed. It is worth saying at the start what the three have in common, because it is not that the numbers are large.

On the human side of the boundary, the problem is never that a term is big. It is that a term is missing.

The one that looks completely harmless

Marking a milestone complete is the action every merchandiser on the desk wanted gone. The evidence arrives as a document: a mill test report, a signed inspection, a receipt. A person opens it, reads one line, and ticks a box. Ninety-six times a month, with 4.4 hours of approval wait each, for what feels like retyping.

It is not retyping. The reason is worth stating carefully, because it is the load-bearing idea of this lesson.

The tick is not a copy of the document. It is a statement by a named person that a state of the world holds: that the fabric really is in the building, in the quantity and the width the order needs. Its whole value to the eleven people downstream is that somebody is standing behind it. A tick that a system produced from a document is not a cheaper version of that statement. It is a different fact wearing the same name: a document arrived that appears to say this. That is a genuinely useful fact, and it is not the one the cutting room is relying on.

You can see the damage in what happens to the other readers. A tick that used to mean checked comes to mean a document was received. Nobody is told, and the people who act on it go on acting on the old meaning. Serdana had one of these before the agent existed, and course 10.3 costs it: a completion marker approved against a planned fabric width, and cloth that arrived narrower.

The repair is a new field, not a new permission.

Automate the preparation, never the decision

Serdana split the row in two. The system now writes a field of its own, evidence received, with the document attached and the relevant line highlighted. A named person still writes the completion. Nothing was automated that was not automatable, and nothing was minted.

Then they measured what that was worth, which is the part most desks never do:

approval wait before        4.4 hours    days of clock lost   0.12
approval wait after         1.2 hours    days of clock lost   0.033
recovered   96 x 0.087 x 214.00                        USD 1,787.33 a month

72.5% of the benefit of automating that row, on a row that was never automated. The reviewer's decision did not change and no attestation moved. The reviewer simply stopped spending three minutes finding the document before making it.

This applies to every row on the human side, and it is the most useful habit in this course:

When an action belongs to a person, automate everything that happens before the decision. The gathering, the highlighting, the cross-checking, the drafting. Leave the decision itself exactly where it is. Most of the clock is in the preparation.

The cost that lands somewhere else

Answering a buyer's status query in the portal fails for a different reason, and the arithmetic in lesson 4 slightly misrepresents it. The USD 2,400.00 in that row is the average of the two wrong answers that actually happened, and neither of those losses landed on Serdana.

Here is one of them. A status answer said the shipment would make the 22nd. Bexholt booked a promotional window on it. The shipment went on the 29th and the buyer took a markdown allowance.

Serdana carried a share of that in the end, in the way suppliers always do. But the decision to remove the approval would have been taken by Serdana, and the exposure it created sits on the buyer's side of the table. A party that cannot see that an answer was produced without a person cannot price the answer.

So this class has a rule that is not arithmetic at all: an action whose cost lands outside your company is not yours alone to move. Serdana's boundary is now a paragraph in the pack it returns to buyers, listing which portal fields are written without a person. Bexholt read it and asked for exactly one change, which was that answers about dates stay human. That conversation cost an hour. It is the reason the row will never be quietly moved by a future manager who has only seen the net.

The wider version of this, what a factory owes its buyers and its workers when it runs these systems and what the contract should say, belongs to course 15.5 and is not settled here.

The action that hides its own errors

The third class is the quietest, and it disqualifies an action before any of the other tests run.

Serdana's plan dates were stored the way most plan dates are stored: one field, overwritten. Write a new date and the previous one is gone. Course 10.3 shows what that costs a calendar. Here it does something more specific.

If a field keeps no history, then after the fact you cannot tell what the agent wrote. So you cannot replay a closed period, so p cannot be measured. And an unmeasurable p does not behave like a large one in a meeting. It behaves like a small one, because the people arguing for autonomy have no number to be held to, and the people arguing against it have no number either.

fields the agent could write                                    9
fields that kept no previous value                              3
hours to make them keep it                                    6.0
rows that moved from unmeasurable to measurable                 2

Six hours. It is the cheapest work in this entire course, and it has to be done first, because until it is done every other number on the page is a guess.

What you own at the end of this lesson

Three tests that come before the arithmetic: is this an attestation, does the cost land on somebody who did not choose the system, and can the error be measured afterwards at all. And the habit that gets most of the value back: automate the preparation, never the decision.

Next: running the thing. The refusals nobody counted, the sample that could never have worked, and when the line has to be redrawn.