Chaos Management

Chaos Management is the other half of the method: how people, priorities and decisions are organised when every developer has full freedom over their work. The price of that freedom is transparency: every document a member works in is visible to every other member of the team. There are no secrets here, for anybody.

Nobody is assigned anything

Work is pulled, never pushed. A trunk has an owner when somebody takes it, and not before. There is no round of allocation at the end of a meeting, and nobody is handed the work that was left over.

That changes what unclaimed work means. When a trunk stays untaken, the method reads it as a finding about the work: it is usually badly described, not actually valuable, or larger than it looks. Handing it to somebody to make the silence stop would destroy the signal.

The spec belongs to the team

Rule 1

The work ships against the spec, or the spec changes first.

Rule 2

A spec changes by team decision. Nobody may change one alone: not the author, not the owner of the trunk, not the person who found the problem.

Rule 1 without rule 2

Anyone who dislikes a constraint edits it and is instantly compliant.

Rule 2 without rule 1

The document is decorative.

Both rules together

The only way past a spec is through the team. Together they are the whole governance model.

So the meetings are load-bearing, not ceremonial. A daily in this method is not a status round, because the map already answers status without anyone speaking. The daily and the review meeting exist to do the two things no document can do alone: give a spec detail it did not need in writing, and ratify a change to one. A meeting that produces neither did not need to happen.

Between meetings, you write it down and keep going. Finding the spec wrong at midnight is not authority to edit it. Record the finding, continue on the parts of the trunk it does not touch, and bring it to the team.

Priority is the order of the cards

Where a board exists, its order carries the priority and nothing else does. A card near the top must be done soon; a card far down can wait. There is no priority field, no label and no number, because a second place to write importance down is a second place for it to be wrong.

Two roles share that, and the split is the whole point:

  • The team decides what is important, in the weekly meeting. Importance is a judgement about the work, so it belongs to the people doing it and is made where they are all present.
  • The chaos master keeps the board clean and ordered. The ordering is clerical, not editorial: they write down the order the team arrived at and keep it legible between meetings. That is why the role can exist without holding power over the work, the same reason a scrum master cannot assign a trunk.

An order somebody disagrees with is not settled by moving the card back. It is raised at the weekly, where importance is decided, and the board follows that meeting rather than leading it.

Who decides what
The team, at the weeklyThe chaos masterThe trunk ownerEach memberAn AI assistant
What is important Decides Writes the order down No part Raises a disagreement at the weekly No part
What a trunk must deliver Decides, and ratifies every change No part Proposes changes Proposes changes Drafts a change, never lands it
Who takes a trunk Settles it when two people want the same one No partNo part Takes it, and writes why No part
How a trunk is builtNo partNo part Decides alone No partNo part
Hours given to the projectNo partNo partNo part Declares their own No part
The status lineNo partNo part Keeps it true No part Keeps it true to the code

Nobody holds a decision that belongs to someone else. The chaos master records the team's order and holds no power over the work.

How a new project starts

Anyone in the team who wants to build an idea writes the specs for it first, then presents them to the team. The team decides where it goes:

  • Ideas, when the team thinks it is good but does not have the time to take on the challenge right now. It is kept with a record of who created it.
  • Todo, when the team thinks it should be built pretty soon.
  • Launched right away, when the team has the time and thinks the idea is brilliant.

Taking on a new project never means dropping another one. We finish the projects we began.

Live projects come first

A security update or a todo on a project that is already live has absolute priority over everything else. People are using those projects, so we cannot accept any delay. Our reputation is at stake.

Ready means checked from several perspectives

We only want to ship high-quality products. In the past, that was ensured by meticulous planning. Our approach lets us launch apps fast, and that could turn against us if we are not paying attention: in theory, only one person writes the specs and leads the main implementation.

So when a product is called ready, it is tested by a non-technical colleague, and at least two developers read its code to find mistakes, and any place where the firm’s principles were ignored. We ensure quality through reiteration and by checking the project from different perspectives: every developer has a speciality and skills in which they excel.

The lead, and a piece of the cake for everyone

The member who takes the lead on an idea becomes its Chaos manager. They are responsible for the whole coordination: between the members, and of the marketing and the financial strategy. In return, the lead receives 50% of the project’s final profit.

At the beginning of the project, the lead states what it needs to launch successfully:

  1. The roles, and how many people each role needs.
  2. The time each person or role should invest.
  3. A percentage for each person or role: the share of the final profit the lead gives them.

The firm matches every percentage the lead gives. A role the lead gives 5% receives 10%: 5% comes out of the lead’s own 50%, and the firm adds another 5% from its part. The whole team should get rich, not just one person: everybody gets a piece of the cake. The firm’s own share never goes below 25%.

Every person and role is filled before the project begins.

Only a member who has taken part in at least five finished projects can lead one, and finished means launched. By then they have a real grasp of how we organise ourselves, and the team has had the time to get to know them and to know whether it wants to work with them.

Leaving a project is an official process, with the reasons written down. If the reasons have to do with the lead, they may be added to the lead’s track record. The goal is not to punish: it is to help the lead understand where their behaviour did not serve the team as a whole.

What leaving does to a share is a provisional rule, and we expect to improve it:

  • The project is not launched yet: the share is lost.
  • The project is launched: the share is kept.
  • The project is launched, and the role also covered maintenance: the share drops by 75%.
  • The role was only maintenance: the share drops by 100%.

For our team, this way of starting projects and sharing profit begins once Vox is live.

Declared hours

Every member states how many hours they intend to give, and everyone can see the number. The member writes it, the member changes it, nobody approves it, and nothing is ever measured against it.

It exists because hours are the missing denominator. Everything else the team can see, what exists in a folder, what a status line says, how long a trunk has been open, is unreadable without it, and people fill the gap by silently inventing a number for each other, usually wrongly and usually unkindly. Four hours a week is a completely respectable thing to give. Four hours a week holding the largest open trunk is a mismatch anybody can see for themselves, with no manager required to point at it.

The only rule attached to it: pull a trunk you can carry at the hours you declared.

Everything a person declares, in one place

Every member has one folder that holds what the team needs to know about how they work: the hours they intend to give, what they are holding, and their own ideas file. The team list is only an index pointing at those folders, so no fact about a person is written down twice, and two copies of one fact never begin to disagree.

Everything in the folder is a declaration. The member writes it, the member changes it, nobody approves it, and nobody writes it for them, an assistant included. A member’s ideas file follows the same rule as the team’s: nothing in it is committed to and nothing in it is scheduled. It exists so an idea stops occupying someone’s head without having to become a proposal first.

Write a fact once

Artefacts are written by many hands: a member, a lead, an assistant in a thread, a module acting on a card. Nobody watches any of them, so the one act they all repeat, writing a fact down, carries three rules. They are local on purpose: each writer sees only the folder they touch, and the order of the whole is the side effect, the way three rules per bird make a flock.

  1. Look before you write. Find where the fact already lives and change it there. Never a second copy.
  2. Write like the place. Take the shape of the folder you land in: its index, its file form, its language.
  3. Land where the subject’s facts already are. The folder holding the most about a subject gets the new fact. A new subject starts its own file there, never a line somewhere else.

What follows without anyone planning it: one copy of every fact, folders a stranger can read, subjects that grow files instead of scattering across chats and to-do lists, and nothing rethought twice, because the lookup is the first step.

The map: no secrets, for anybody

Everybody has to be able to see what exists without asking for it. That is what the freedom is paid with. What a map has to provide, stated as requirements rather than as a product:

  1. Publishing costs nothing. You keep working exactly as before, in documents the whole team can see. Anything that asks a person to upload, tag or summarise is a report, and reports stop getting written in week three.
  2. Every document, visible to every member. What one of us works in, all of us can open. What is never recorded is the person: no activity trail, no read receipts, no “last seen editing”.
  3. Everyone sees everyone. A view only a lead can open is supervision wearing a map’s clothes.
  4. It shows the interval nothing else shows: what exists before it is finished.
What each kind of tracking can see during one trunk
HeaderVersion controlThe map
Trunk taken, claim written seennot seenseen
First notes and drafts The interval only the map showsnot seennot seenseen
A branch nobody has opened a review for not seennot seenseen
A week stuck on one problem not seennot seenseen
First commits pushed not seenseenseen
Status line: first slice built seenseenseen
Done seenseenseen

An illustration, not data. The header changes when a trunk is taken and when something is finished, and commits arrive once work is pushed. The map shows what exists in between.

A version control history is a partial map, and the part it misses is the important one. Commits answer what was finished and pushed, which the status line was going to say anyway. The drafts, the notes, the branch nobody has opened a review for, the week where somebody is stuck: that interval is only visible on a map.

A team of many, working as one.

How far along are we?

When an assistant is asked, it answers from artefacts: the status line, the commits, what exists in the folder, and what the owner said they set out to do. When the artefacts do not say, the answer is “I do not know”, not an inference. It never turns activity into a number that reads as a score.

What the method does not have

Their absence is the design, not an oversight.

  • Estimates
  • Story points
  • Velocity
  • Burn-down charts
  • An assignee field
  • A per-person activity feed
  • A ticket state machine
  • A sprint boundary work must fit inside

What it costs

The method is weak in places, and says so. This list is supposed to shrink.

  • Specs bloat. A spec that cannot be re-read in one sitting is write-only, and its invariants hide inside it instead of being stated by it.
  • Specs drift. When the work lands and nobody updates the spec, the code quietly wins. The status line is the cheap repair.
  • A wrong trunk burns longer before anyone notices. Stories give early exit points; trunks do not. The substitute is the map: something visible on disk early in the trunk, not a slice of it.
  • Waiting for the team is a real delay. A spec found wrong on Monday can cost days if the meeting is on Thursday. That is the price of the team owning the spec, and it is paid knowingly: the alternative is specs that mean nothing.
  • The header says who owns a trunk, not how far along it is. That interval is exactly what the map is for, and why the method is incomplete without one. A header that has not changed in six weeks is the one thing the method genuinely cannot read for you.