留四个，恶者辄去毋令败群

Eighteenth of March, three in the afternoon, third day back. On the table, one sheet: eight rows, five columns. Eight leaders; bags sold in 2030, refund rate, repeat rate, days from group-open to delivery, and a notes column.

I have had this table since December. Today I printed it again, because Jiang's sentence has come due. In July 2029 she said: find eight, keep four. I took that for natural attrition. In a whole year not one of the eight left.

She read it once and said: who are you cutting? I haven't decided. You have. Pushing it back: if you hadn't, you wouldn't have printed this today.

Bu Shi came in the evening after she'd gone — fifty or so, very coarse cloth, straw sandals, wool stuck to his clothes, not embroidered on, a long staff with a slight crook at the top. He looked at the table. You've sorted it five times. Yes. All five times you were looking for the same thing. Which is? A reason not to cut the one who sells the most. Setting the staff against the wall: Bu Shi.

On the sheep: I started herding in my teens. My parents died and there was a younger brother; when he grew up I left. With what? A hundred-odd sheep. The fields, the house, the goods, I gave him. Why? Sheep breed. Flatly. Fields don't. And: a dozen years in the hills, a hundred-odd became more than a thousand. And: for a dozen years I looked at the same question every day. Which was? Which one has to be taken out.

He can't read the characters but he reads figures. Staff-tip on row three — Hé. This one. She sells 480, second highest. I know. The tip moved right, to the 9.2%. That number is three times the other seven. I said a high refund rate doesn't make somebody a bad person. I didn't say she was bad. Immediately. A sick sheep isn't a bad sheep. It's suffering too. And, and I picked the pen up: the problem isn't in her. It's in who she's standing next to.

On the sentence itself: keep to the hours of rising and resting; the bad ones you take out at once, so they don't spoil the flock. What does bad mean? You later people read that character as wicked. Read it as wicked and the sentence becomes: drive out bad men. That is not what I meant. Raising the staff: the weight is in the last four characters. So they do not spoil the flock. And: I take one out not because it is bad, but because if it stays the flock goes with it. And, three times into the book: what you are judging is not the animal. It is what the animal is doing to the flock.

Then a specific question. Where does that 9.2 come from? Her 480 bags produced 44 claims: 29 "faint", 9 "sticky", 6 other. Does anybody else get "faint"? The other seven together: 1,930 bags, 17 claims of "faint." Same goods? Same lot, out of the store on the same day. Staff-tip to the last column. Eleven days. This.

And then the part I have no answer to. Her practice is to order a large quantity, keep it at home and hand it out batch by batch; eleven days from her receiving it to the last customer getting it; on a Chengdu balcony, twenty-seven or twenty-eight degrees in the afternoon in May and June. I raised it with her once last year. What I said was: "Hé jie, better to hand it all out at once." She said "fine" and did not change it, and I did not raise it again, because she sold a lot. Bu Shi: you told her once. I did. What you told her was "better." And: "better" is not a rule. And, and I drew a heavy line under it: you never gave her one she could keep. — M.

Memo to file. Chengdu, 18–20 March 2031. Cutting four leaders of eight.

1. The table, 2030. Xiao Lei 620 / 3.1% / 41% / 4d. Guo laoshi 410 / 2.8% / 38% / 5d. Hé 480 / 9.2% / 19% / 11d. Chen 180 / 7.9% / 52% / 3d. Wang 300 / 4.0% / 30% / 6d. Liu 210 / 3.5% / 26% / 7d. Zhang 120 / 2.0% / 11% / 14d. Zhou 90 / 1.5% / 8% / 16d. Total 2,410.

2. Cutting on volume would have cut Chen, who is eighth-but-two on bags, first on repeat purchase (52%), and who invented and keeps the what I remember column that produced the single most useful correlation I own (70 mm of rain, 9 parcels, 4 sticky). On the ledger alone she is a loss: 180 bags earns me ¥1,935 and I pay her ¥3,600 a year. Any cut that runs down the volume column removes my only instrument.

3. The criterion, corrected. Bu Shi's sentence is not remove the bad. The weight is in "so they do not spoil the flock." You are not judging the animal; you are judging what it is doing to the flock. Applied: Hé's 480 bags produced 29 "faint" claims. The other 1,930 bags across seven leaders produced 17. Same lot, same despatch day. Her 11-day hold on a 27–28°C balcony is not her problem, it is a contamination source in every batch statistic I compute.

4. Decision, and its price. Keep: Xiao Lei, Guo, Chen, Wang — 1,510 bags in 2030. Cut: Hé, Liu, Zhang, Zhou — 900 bags. 900 × ¥10.75 = ¥9,675 of margin removed for 2031, against ¥62,000 in the bank and ¥100,000 a year of burn. First time in three years I have deliberately cut my own revenue, and I did it knowing I cannot replace it.

5. The phone calls. Liu, Zhang and Zhou took about a minute each. Zhang asked whether it was because she sold too little; I said yes, without dressing it, and she said she understood. Hé's call ran nineteen minutes. She said, evenly, "I sell more than any of them." — except Xiao Lei — "then why are you cutting me?" Two true answers were available: your refund rate is three times everyone's (true, and lands as an accusation about her), and you hold it eleven days (true, and the cause). I gave the second, with the numbers: eleven days, 27–28°C, 29 against 17 on nearly a quarter of the volume. Long silence. Then, not loudly and not angrily: "你早说啊 — you should have said."

6. And she is right, and this is the actual finding of the week. — What I said last year was "better to hand it all out at once." No day count. No temperature. No consequence. A suggestion, and suggestions may be declined — and I phrased it as a suggestion because she sold a lot and I did not want to offend her. — Her closing offer, unprompted: "If you'd told me last year that it starts going faint on day eight, I'd have changed it. I've got a chest freezer on that balcony doing nothing." — Liu 7 days, Zhang 14, Zhou 16. Two of the people I cut hold stock longer than Hé, and both joined in July 2029, and nobody ever told them anything about it.

7. The leader handbook, re-read this evening. Twelve pages. How to open a group, how to collect payment, how commission is computed, how to write the post. Two lines in twelve pages concern the goods themselves: store in a cool place, avoid direct sunlight; and it is suggested you distribute promptly after receipt. Where is cool? How many days is promptly? I ran the experiment in May 2029. I know the answer: above 27°C, from about day eight, three in ten repeat buyers report "faint." I have held that number for nearly two years and never once put it in the handbook.

8. Which reframes the cut. Of the four people I removed, at least two were removed for failing a rule that was never written down — and one of them told me she would have complied on the day she was told. That is not culling. That is me charging other people for my own omission. I am not reversing the cut. Hé's contamination was real and this year's numbers have to be clean. But the count stands: 900 bags of margin, and a share of it was mine to lose by not writing two paragraphs.

9. Chen, by message tonight, and she went further than Bu Shi did. — "Hé asked me last month how many days it keeps." What did you tell her? "I said I don't know, I hand it out in three days." And: "She said she'd like three days too, but she orders a big lot and can't clear it in three." — And then: "Marco, those eight lines you print on the bag are for the buyers." And: "You've never printed one line for us."

10. Standing action for next week, unfunded and free: a written charter for the four remaining, with numbers in it, before the May run, not after it. Everything I know about day counts, temperature, storage, batch handling and what happens when they are broken. The measure of it is Bu Shi's: it has to be a thing a person can keep, not a thing a person is advised to consider.

"What you are judging is not the animal. It is what the animal is doing to the flock."
He had left home with a hundred-odd sheep and given his younger brother the fields, the house and everything else, on a piece of reasoning he saw no need to expand: sheep breed; fields don't. A dozen years in the hills turned the hundred into more than a thousand, and asked what he had spent those years doing, he described it as one question repeated daily. Which one has to be taken out.
The sentence he is remembered for was given to an emperor who had stopped to admire the imperial flock and got a lesson in government instead. It is not only sheep. Governing people is the same. Keep to the hours of rising and resting; the bad ones you take out at once, so that they do not spoil the flock.
He was impatient with how the middle clause has been read since. You later people read that character as wicked. Read it as wicked and the sentence becomes: drive out bad men. That is not what I meant. The weight, he insisted, is in the four characters at the end. I take one out not because it is bad, but because if it stays the flock goes with it.
Shown a table of eight sellers and asked which to remove, he put his staff on the row with the second-highest sales, and when told so, said he knew. Then he moved the tip to that row's complaint rate — three times everybody else's — and, told that a high complaint rate does not make a person bad, agreed at once. A sick sheep isn't a bad sheep. It's suffering too. And: the problem isn't in her. It's in who she's standing next to.
The cause turned out to be eleven days of storage on a warm balcony, and the man who had come to make the cut had raised it with her once, a year earlier, in these words: better to hand it all out at once. She had said fine and carried on, and he had not raised it again, because she sold a lot.
You told her once. What you told her was "better." "Better" is not a rule.
You never gave her one she could keep.