---
title: "Chapter 7: The Control Room"
book: "Collapse or Collapse"
author: "Vinay Pasricha"
url: https://vinaypasricha.com/books/collapse-or-collapse/chapter-7-the-control-room
status: "On Kindle: https://www.amazon.com/dp/B0HNL53J5R. The full text is shared on vinaypasricha.com with the author's permission."
---

# Chapter 7: The Control Room

The master’s room follows the same rule as the cockpit: one face, and nothing shown until asked. The difference is the floor.

A master holds one Outcome and every vector running on it, and the number of vectors may be three or may be more than a thousand.

That sentence would have been absurd to every management thinker of the last century, and they had arithmetic on their side. In 1933 a Lithuanian-born management consultant working in Paris, V. A. Graicunas, counted what a manager actually has to hold in her head: not only her relationship with each subordinate, but the relationships between them, and with every group of them that might act together. The count does not grow with the team. It explodes. With four subordinates a manager holds 44 relationships; with five, 100; with six, 222; with twelve, more than twenty-four thousand.[^37] Lyndall Urwick, the most influential management writer of his generation, turned the arithmetic into a rule that every business school still teaches in some form: no superior can directly supervise more than five or, at the most, six people whose work interlocks.[^38]

They were right, for the world they were in. A manager supervised people, and people’s work had to be watched, coordinated and carried across, and that cost attention that grew faster than the team. The whole pyramid of the old company was built on that limit: six people under each manager, six managers under each director, and so on up, until a company of ten thousand had five or six layers between the person at the top and the work.

The control room breaks the arithmetic, and it breaks it by changing what the master holds. She does not supervise people whose work interlocks. The interlocking is inside the chains, where the agents carry it, and the vectors do not need her to coordinate them with each other. What she holds is the state of each vector, shown as a colour, and the few things that need a human. Graicunas’s relationships do not arise, because there is no work passing between her pairs for her to carry. She does not read a thousand vectors. She reads the colour of a floor, and walks to the red.

## The floor

When she asks to see her Outcome, the floor becomes it: one tile of light for every vector, hundreds or a thousand of them stretching away like a city seen from above at night. Most are pale and steady, which means they are flowing. The ones with a warning are amber. The ones that need her now are red, and they are always nearest her feet, because the floor sorts itself. She does not scan. She reads the colour of the whole, which tells her the health of the Outcome in a glance, and then she reads the colour at her feet, which tells her where to go.

![Picture 8: The control room](https://vinaypasricha.com/assets/images/collapse-or-collapse/pictures/08-the-control-room-1600.webp)

*Picture 8. Key. What: the control room; the floor is the Outcome. Who: the master (human, left) and the quartermaster of the red vector nearest her feet (agent, on the disc); it belongs to that vector, not to her. What is happening: every tile is a vector; pale is flowing, amber is a warning, red needs her now, and the red tiles have already moved nearest her. She is about to walk to them.*

She can go as deep as she likes, until there is nothing left to see. She steps onto any tile and she is in that vector’s cockpit, where its own quartermaster explains the situation and shows her its reports, exactly as it would show the aligner. From there she can ping the aligner or the tuner of that vector, by message or by voice, and check in. Nothing is hidden from her at any level. She chooses how far down to look, and most days she does not need to look far, because the vectors’ own quartermasters have already looked, and the floor is what they report.

What makes a tile amber or red is not one measure but several, each with its own colour inside the tile: results against target, the attention of the pair, a technical fault, cost running ahead of value, the satisfaction of the vector’s customers, and anything else the Outcome has learned to watch. Each of those colours is reported by an agent inside the vector itself: the quartermaster for the result and for the pair’s attention, the agent maker for a fault, the vector’s accounts agent for cost, its client agent for the customer. The floor is the vectors speaking, and nothing stands between them and her. The tile takes the colour of the worst thing inside it, so a red inside is red on the floor, and the master steps in to find which of the several it is. This is the rule of the tile, and it will reappear one level up, in the command centre, because it is the right way for any whole to show the state of its parts: the outside colour is never calmer than the worst thing inside. A rule that softened the colour on its way up would hide exactly what the floor exists to show.

## Nobody beside her

Nothing works for the master except the floor. There is no staff beside her, human or agent: no head of her Outcome, no HR, no quality head, no accountant, no technologist. She is the only human above the pairs in her Outcome, whatever its size; her standby is another master; and the agents she sees are the agents of her vectors, one at a time, when she steps onto their tiles.

Every function a division head once kept at her elbow is in the vectors instead, at the step where it is needed, and each is there by the one test this book applies to every piece of work: a skill is an agent, inside the vector that needs it; a result is an Outcome, with a master of its own. Nothing is a department, and nothing sits beside the master. Where each of the old functions went is worth setting out, because a reader who has run a division will look for them.

**Quality** is an inspector agent in every chain. It checks every output of its vector before the output passes a gate, around the clock, against the playbook, and is built not to miss anything; when it does, its agent maker rebuilds it. It may hold anything in its own vector. When it holds an output that the quartermaster passed, the output waits; either of the pair may appeal to the master, who decides. This is the only dispute inside an Outcome, and it has one court. What the inspector finds is one of the colours inside the tile, so the master sees quality on the floor without a quality department to report it to her.

**Cost** is an accounts agent in every chain. It keeps the cost of its vector and the value the vector delivers, and shows the difference as a colour in the tile. It pays the vector’s own costs, a freelancer, a tool, an agency that runs a step, within the limit the playbook sets for it; above the limit it asks the master, because money is an irreversible gate. It pays the aligner and the tuner by the formula set out in the chapter on pay, on the vector’s own result. The Outcome’s budget is the sum of its vectors’ budgets, and the master holds it against the one budget the Keeper gave her. There are no internal prices between Outcomes.

**People** are the vector’s own concern. A new aligner or tuner is found by an agent that runs the selection from application to offer and signed for by the vector’s tuner, as the chapter on selecting describes, and is trained by the vector itself, which teaches her the work from the first morning. An agent that has failed is decommissioned by the vector’s agent maker, which can restore it, so there is no gate. A human cannot be restored, so letting a human go is an irreversible gate, and the only hand on it is the master’s. The evidence for that decision comes from the work: the quartermaster measures attention from the gates answered and the reviews kept, and shows the tile amber weeks before the pair would say anything. The chapter on attention sets out what happens then.

**Technology** is the agent maker’s. Each vector’s agent maker watches for new tools, new models and new versions as they appear, and proposes each upgrade to its tuner, who accepts or refuses it for that vector, because the agent maker alone knows the local customisations of its vector and must fit the new technology around them. A better way found in one vector reaches the others through the playbook and only through the playbook: the master writes it in, every agent maker reads it, and each installs it in its own chain. Nothing spreads across vectors by any other path.

**The health of the Outcome** is the floor. There is no agent whose job is to watch the vectors for her, because the vectors watch themselves and report, and a watcher above them would be the first tile of a new pyramid. The next section says why.

Her controls are three: question a vector, pause it, stop it.

And one rule above all the others. The agents may read anything inside the engine and nothing outside it. That is the boundary, and it is absolute: no agent of the company reads an aligner’s private messages, her applications elsewhere, or her life. What the engine shows is enough. An aligner who is leaving stops answering gates as fast, lets the review streak break, and lets the quality of her judgement calls drift; her vector’s quartermaster reads those signals and shows the master the tile amber weeks in advance, so that a standby is ready before the gap opens. The prediction comes from the work, never from watching the person. I will say more about this in the chapter on attention, because it is the line between a company that measures and a company that surveils, and the doctrine is on one side of it.

## Why there is nobody beside her

People ask why the master has no staff, and the answer is the same as the answer to why the aligner has no team.

A human beside the master would be a layer. Not at first; at first she would be an assistant, a chief of staff, a deputy. But a deputy who reads the floor for the master becomes the person the aligners report to, and then the person who decides what the master sees, and then a manager, and the pyramid is growing again from the top. The master is alone above her pairs not because she is heroic but because the moment she is not alone, the company has a fourth layer, and four is the first step back to seven.

An agent beside the master would be a department. That is the subtler mistake, and the easier one to make, because an agent does not look like a layer. An agent that checks quality for every vector from outside them is a quality department with no people in it: it holds the vectors’ outputs, grows its own rules, needs its own budget, and on the day it needs a human to own it, a function has returned to the top of the Outcome with a hold over every result and responsibility for none. The same is true of an agent that keeps the Outcome’s accounts, or its people, or its technology. Each would be a skill standing outside the work it serves, which is the definition of a department. So each stands inside the work instead, in every vector, at the step where it is needed, behind a gate, and the master sees it as a colour on the floor. An Outcome has no functions. It has vectors, and a master.

Her standby is another master, and that is also not an accident. (In the small company of chapter 15, where there is only one master, her standby is a backup who has shared her room.) The only human who can hold an Outcome in a master’s absence is someone who already holds one, because holding an Outcome is a craft, and the craft is the same across Outcomes even where the chains differ. Two masters, each the other’s standby, each with the other’s playbook open, is the whole of the company’s insurance against a master’s absence, and it costs nothing, because both of them were already there.

## The rules of the control room

A master in her first month meets the same questions every master meets: how to split a vector that has grown, when to give an aligner a second one, how a new pair is trained, what to do when an inspector and a pair disagree, how she is paid, who stands by for her, what happens when her whole Outcome turns red. The doctrine answers all of them in advance, in eighteen rules. They are set out in one line each at the end of this chapter, for the master who wants them beside her. Most of them are easier to understand in use than in a list, so before the list, here is a control room in use.

## A morning at GoodSpace

The master of GoodSpace’s largest Outcome is a man named Mudassir, who was its delivery head before the word master existed in the company. His Outcome is the one the last chapter visited from the inside: a job description goes in, and an invoice comes out. It runs as ten vectors, each with an aligner and a tuner and about fifty open mandates, which makes some five hundred roles that clients are waiting for, four hundred or so of them live in any week, and between eighty and a hundred and twenty closed in a month. The mornings below are composites of several, with the clients and candidates changed; the decisions are the kind he makes every week.

Ten vectors is a small floor. It is not a thousand tiles; it is ten, and inside each of them the mandates show as smaller tiles of their own, so that when he looks down he sees ten pale squares, each made of fifty smaller ones, and the few that are not pale. On this morning three vectors’ tiles are red, one is amber, and the rest are pale. He walks to the red.

The first is a mandate that has failed the test the pairs set themselves. Three batches of candidates have gone to the client, interviews have been held, feedback has come back, and there is no selection. The aligner has already sat down with her quartermaster to find the leak, as the vector’s own rule requires, and has sent the question up, because the answer is not hers to give. The feedback trail is plain when he reads it. The client rejected the first batch for experience, the second for salary, and the third for reasons that contradict the first. The findability report that opened the mandate scored it a six; the client’s own behaviour has since made it a nine. Mudassir does not send a fourth batch. He has three choices, and he takes the hardest: he calls the client’s hiring head himself, puts the three rounds of feedback side by side, and asks which of the three people the client actually wants. It is the conversation the chain cannot have, because it is a conversation about the client’s own indecision, and only a person the client trusts can hold a mirror to it. The second red mandate is simpler. Its client has gone silent for three weeks after promising feedback, and Mudassir releases it to hold and frees the pair’s time for mandates that are moving.

The second thing is capacity. One vector carries fifty-eight mandates this week, because a large client opened eight roles at once, and its tile has gone amber on load; another carries forty-one, because a client paused hiring. He steps onto the heavy tile, and its quartermaster shows him the eight new roles and which pair in the Outcome has worked that client’s roles before: the pair on the light vector. He moves six mandates across, the two agent makers carry the mandates’ history between the chains, and by the afternoon both vectors are near fifty. This is a rule of the control room in action: the vector reports its load, the master moves the work, the agent makers carry the record.

The third is a colour, not a mandate. One client’s own tile, the customer’s colour that the floor shows beside the Outcome’s, has been drifting from pale to amber for a fortnight. Nothing has failed. But interviews are being rescheduled, feedback is slower, and the client’s last two messages were shorter than its first twenty. A customer leaving is the one thing in his Outcome he cannot fix after the fact, so he does not wait for a red. He puts a call in the diary for the afternoon, not to sell anything, but to ask whether something has changed at the client. It usually has.

The last is a dispute. The inspector in one vector held a candidate that the vector’s tuner had signed out the evening before: the candidate’s skill assessment was nine months old, and the playbook says a skill assessment older than six months is retaken before a candidate is forwarded. The tuner appealed. She had spoken to the candidate for twenty minutes, he was the strongest on the mandate, and the client wanted him seen this week. Mudassir reads both sides and upholds the hold. The rule is in the playbook for a reason, and he wrote it himself; a rule that bends for the strongest candidate is not a rule. The candidate takes the assessment that afternoon and, if he passes, goes to the client the next morning, one day later than the tuner wanted.

That is the morning. Two red mandates, one capacity move, one customer drifting, one dispute. Nothing else on ten vectors and four hundred live mandates needed him, and he knows it did not need him because the rest of the floor is pale. Not all of the agents inside his vectors are yet as complete as this chapter describes them; where one is missing, a report or a person still fills the gap, and the next chapter says so plainly. But the shape of his day is already the shape of the doctrine: he reads colours, walks to the red, and makes the few decisions that a human must make.

Will the same room hold a thousand vectors? The honest answer is that GoodSpace has held ten, and a thousand is the doctrine’s claim, not its record. But the arithmetic is on the claim’s side, for the opposite reason that it was on Graicunas’s. His relationships multiplied with every person added, because every person’s work had to be coordinated with every other’s. The control room’s load grows only with the exceptions, the reds that reach the master, and the exceptions per vector fall as the playbook improves, because every red the master resolves becomes a rule the vectors follow. A master with ten vectors and five reds a morning, and a playbook that halves its reds every year, will hold a hundred vectors with the same five reds in a few years, and a thousand after that. The limit is not the number of tiles. It is the number of things that turn red, and that is a number the master herself drives down. And if a floor outgrows her before the playbook catches up, if the reds reaching her week after week are more than her review can clear, the doctrine does not give her a staff. The Keeper splits the Outcome in two, each whole, each with its own master and a copy of the playbook, as a vector is split when its pair can no longer hold its gates. A master’s capacity is a limit to design around, not a reason to build a department beside her.

The next chapter goes up one more level, to the man who holds every control room at once.

## The eighteen rules of the control room

1. **The playbook.** Every Outcome carries a written design (chain, gates, authority limits, standards, health thresholds); the master owns it, every agent maker reads it, and a new vector opens from it in a day.
2. **Splitting and merging.** The master may split a vector that has grown or merge two that have thinned, without the Keeper; the agent maker clones itself into the child.
3. **Learning across vectors.** A better way found in one vector enters the playbook; every agent maker reads the playbook and installs the change in its own chain. Nothing spreads except through the playbook.
4. **The daily review.** Once a day, at an hour she fixes, she reads the floor in a fixed order: red vectors first, then the pairs’ attention, then quality, cost and the customers’ colour, each only where it needs her.
5. **Capacity.** A vector’s quartermaster reports its load; the master moves work between vectors; the agent makers carry the record across.
6. **Onboarding a pair.** A new aligner or tuner spends a fortnight under supervision, on a trial vector or co-signed on a live one, before holding a live vector alone.
7. **Agents and humans.** An agent maker may decommission an agent in its own vector, because an agent can be restored; only the master may let a human go, because a human cannot.
8. **The customer’s colour.** The floor shows the Outcome’s customers and their satisfaction as a colour of its own.
9. **The audit trail.** Every decision by every agent and human is logged and replayable; an agent that cannot show why it acted is stopped until it can.
10. **Quality disputes.** A vector’s inspector may hold any output of its vector against the playbook; either of the pair may appeal; only the master may overrule.
11. **The master’s standby.** Another master, holding a different Outcome, stands by for her, with her playbook always open.
12. **Who watches the master.** Her review, her speed on reds and the decisions waiting on her are her Outcome’s colour in the command centre.
13. **Targets.** The Keeper gives one result and one budget; the master sets everything beneath them.
14. **Authority.** Within budget and playbook she needs nobody’s permission; a sum beyond the budget or an act outside the playbook goes to the Keeper.
15. **Pay.** A fixed base at the intention premium, and a share that moves with the Outcome’s result against target.
16. **How a master is made.** Chosen by the Keeper from aligners who have run two or more vectors and stood tuner for others, after a month beside the outgoing master. At the collapse, when there are no aligners yet, the first masters are chosen by the four layers (chapter 13).
17. **When the Outcome goes red.** Within twenty-four hours the Keeper calls the master and her standby; the Keeper decides a new target, a new playbook or a new master. The Outcome is never paused and never runs without a master.
18. **When the floor outgrows her.** If the reds reaching the master exceed what her daily review can clear, week after week, the Keeper splits the Outcome in two, each whole and each with a master. The floor is never given a staff.

[^37]: V. A. Graicunas, “Relationship in Organization,” *Bulletin of the International Management Institute*, March 1933; reprinted in Luther Gulick and L. F. Urwick (eds.), *Papers on the Science of Administration* (Institute of Public Administration, Columbia University, 1937). Graicunas counted direct single, direct group and cross relationships; the maximum total for n subordinates is n(2^(n−1) + n − 1): 44 for four, 100 for five, 222 for six, and 24,708 for twelve.

[^38]: L. F. Urwick, *Scientific Principles and Organization* (American Management Association, 1938), and “The Manager’s Span of Control,” *Harvard Business Review*, May–June 1956. Urwick’s formulation was that no superior can supervise directly the work of more than five or, at the most, six subordinates whose work interlocks.