The standing map, and what would have to change

What Rina actually wanted

Not a ranking. She wanted a document that lets somebody re-run the argument next February and reach the same answer, or a different one, on purpose, without starting again.

A ranking cannot do that. It has thrown away the two things a reader needs: what each number is a fact about, and how confident anybody is entitled to be that it is still true. So the map has four columns before it has any content at all.

• Unit — country, cluster or factory, from lesson 1's ten-second test.
• Evidence — MEASURED, QUOTED, READ or UNKNOWN.
• Dated — the day the evidence was taken, not the day the sheet was typed.
• Horizon — how long this row would take to change, and who would change it.

The evidence column, which is the one people resist

Four classes, and the distinction is not fussiness.

MEASURED — your own record. Your goods-received dates, your payroll, your quality log, your win-and-loss sheet. It has a date, and you can produce the underlying rows.

QUOTED — somebody told you, in writing, for a purpose. An FOB price, a freight rate, a lead time, a mill's minimum. It has an expiry, whether or not one is printed on it. A quoted lead time is not a measurement. It is a hope, and the comparison rewards whoever is most optimistic on the telephone.

READ — you saw it somewhere. An article, a slide deck, a supplier's own presentation, a number somebody said in a meeting two years ago that has been repeated since.

UNKNOWN — you do not know.

A READ row may not carry weight in a decision. Convert it — measure it, or get it quoted by somebody accountable — or move it to UNKNOWN and leave it visible. This is not a counsel of perfection. An UNKNOWN in a map is useful, because it is a task with a name against it. A READ row dressed as a fact is the only one of the four that can do damage, because it looks finished.

The temptation it protects you from is a national statistic. Course 26.6 gives the sharpest reason to resist one: whoever is across the table has better numbers than you have, and they are about you rather than about your country. A country average cannot survive contact with four quotes and four despatch records. Reaching for one tells the other side which of the two you would rather discuss.

Horizons, and the rule that follows from them

The horizon column is the one this course has been building towards. Every row can change. They change at wildly different speeds, and by wildly different hands.

Read the column and the review rule writes itself: review a row on the shorter of its horizon and the shelf life of its evidence. A row that can change in three months and was measured eighteen months ago is not a fact, whatever it says.

Course 26.1 gives the other half of the same rule, and the two work together. Its version is fragility — how far a number has to move before the answer flips. That settles which rows are worth watching. Horizon settles how often to look. A fragile row with a long horizon can be left alone for years. A robust row with a three-month horizon still needs checking, because robustness is a property of today's gap, and the gap is what horizons close.

The mistake nobody made

Norforth's map was built well in year one and re-approved in year two. The review was a real review: two people, half a day, every row read, arithmetic checked, internal consistency confirmed. Nothing was found, and the map was signed.

Three rows had changed. None of them was caught, because each of the three was checked for the wrong thing.

• A preference into Ambernay had been narrowed in scope. A country row. The review looked at it, and it still said "duty-free" — which was true for two of the four categories Norforth buys, and no longer true for the third.
• A dyehouse had opened in Marendal. A cluster row. Nobody re-asks a cluster question inside a year, because clusters are famously slow.
• Ferrand's on-time record had fallen from 91% to 68% after it took a large programme from somebody else. A factory row, and the map's number was MEASURED — measured in month three of year one, twenty-one months earlier, against a horizon of a few months.

The third one cost money. Norforth placed 24,000 units of a fashion drop with Ferrand, on a reliability figure that was twenty-one months old, and did not know it. The map had no date column and no horizon column. So the number that was most out of date looked exactly like the number checked last week.

Both reviewers did their job. What they had was a map that could only be checked for consistency, and it was perfectly consistent. A document that cannot be wrong in a way you can detect is not a safe document. It is a confident one.

Where the changes actually were

Over three years Norforth logged every change it made to the map: 31 of them.

93.5% of the changes were in rows that are not country rows. And the country rows are the ones the annual meeting spends its time on, because they are the ones with names attached that everyone has an opinion about.

This is not an argument that country rows do not matter. The two that changed mattered a great deal: a duty treatment is worth more per change than anything else on the sheet. It is an argument about attention. The rows that move are the rows closest to you, they have the shortest horizons, and they are the ones people believe they already know.

What the map may never do

Two things, and both turn a good map into a bad decision.

It may not become an approved-country list. Course 26.1 puts the reason plainly: an origin is good or bad into a market, for a product, at a quantity, this season. So a standing list of permitted countries is a decision to be wrong somewhere. The map narrows where you look. It never chooses.

It may not fill an empty cell. A blank is honest. A zero is a claim. A plausible number written in to complete a row is worse than either, because the next reader cannot tell it from a measured one. Write UNKNOWN, and the name of the person who could answer it.

The last thing to write down

At the bottom of the page, under everything else, put the sentence that says what would change the answer. Write it for the row with the shortest horizon and the largest weight, whichever that turns out to be this year.

For Norforth in year three it was one line: if Ferrand's on-time record does not recover to 85% by March, the newness drops move and the carry-over stays. One row, one threshold, one date, one consequence. Everything else on the page exists so that this sentence can be written honestly.

Act as a sourcing analyst auditing a document rather than writing one. I will paste my origin map [PASTE IT]. For every row, fill in four columns and change nothing else. UNIT: country, cluster or factory. EVIDENCE: MEASURED if it came from my own records and I can produce the underlying rows, QUOTED if somebody told me in writing for a purpose, READ if I saw it somewhere, UNKNOWN if I do not actually know — and where I have not told you which, ask me rather than guessing. DATED: the day the evidence was taken, not the day the sheet was typed; if I have not given you one, write NO DATE and treat the row as expired. HORIZON: how long this row would take to change and who would change it, using country rows of years, cluster rows of one to several years, and factory rows of months. Then do four things. Flag every row whose age exceeds the shorter of its horizon and the shelf life of its evidence, and sort those by the weight they carry. List every READ row and tell me it may not carry weight in a decision until it is measured or quoted by somebody accountable. Tell me which single row, if it moved, would change my current placement, and how far it would have to move. Finally, write one sentence for the foot of the page in this form: if [ROW] does not reach [THRESHOLD] by [DATE], then [CONSEQUENCE]. Do not fill an empty cell with a plausible number, and do not use any national statistic anywhere.